Spaces:
Running
on
T4
Running
on
T4
Commit
·
2d42726
1
Parent(s):
2d134d2
gradio files are synced with the github repo
Browse files- .gitattributes +0 -34
- README.md +2 -2
- app.py +36 -18
- configs/latent-diffusion/gqa-inpaint-ldm-vq-f8-256x256.yaml +122 -0
- constants.py +1 -1
- ldm/__init__.py +0 -0
- ldm/lr_scheduler.py +98 -0
- ldm/models/autoencoder.py +426 -0
- ldm/models/diffusion/ddim.py +204 -0
- ldm/models/diffusion/ddpm.py +1062 -0
- ldm/modules/attention.py +271 -0
- ldm/modules/diffusionmodules/model.py +836 -0
- ldm/modules/diffusionmodules/openaimodel.py +973 -0
- ldm/modules/diffusionmodules/util.py +270 -0
- ldm/modules/distributions/distributions.py +92 -0
- ldm/modules/ema.py +76 -0
- ldm/modules/encoders/modules.py +52 -0
- ldm/modules/losses/contperceptual.py +111 -0
- ldm/modules/losses/vqperceptual.py +167 -0
- ldm/modules/x_transformer.py +641 -0
- ldm/util.py +211 -0
- requirements.txt +2 -1
- utils.py +7 -57
.gitattributes
DELETED
@@ -1,34 +0,0 @@
|
|
1 |
-
*.7z filter=lfs diff=lfs merge=lfs -text
|
2 |
-
*.arrow filter=lfs diff=lfs merge=lfs -text
|
3 |
-
*.bin filter=lfs diff=lfs merge=lfs -text
|
4 |
-
*.bz2 filter=lfs diff=lfs merge=lfs -text
|
5 |
-
*.ckpt filter=lfs diff=lfs merge=lfs -text
|
6 |
-
*.ftz filter=lfs diff=lfs merge=lfs -text
|
7 |
-
*.gz filter=lfs diff=lfs merge=lfs -text
|
8 |
-
*.h5 filter=lfs diff=lfs merge=lfs -text
|
9 |
-
*.joblib filter=lfs diff=lfs merge=lfs -text
|
10 |
-
*.lfs.* filter=lfs diff=lfs merge=lfs -text
|
11 |
-
*.mlmodel filter=lfs diff=lfs merge=lfs -text
|
12 |
-
*.model filter=lfs diff=lfs merge=lfs -text
|
13 |
-
*.msgpack filter=lfs diff=lfs merge=lfs -text
|
14 |
-
*.npy filter=lfs diff=lfs merge=lfs -text
|
15 |
-
*.npz filter=lfs diff=lfs merge=lfs -text
|
16 |
-
*.onnx filter=lfs diff=lfs merge=lfs -text
|
17 |
-
*.ot filter=lfs diff=lfs merge=lfs -text
|
18 |
-
*.parquet filter=lfs diff=lfs merge=lfs -text
|
19 |
-
*.pb filter=lfs diff=lfs merge=lfs -text
|
20 |
-
*.pickle filter=lfs diff=lfs merge=lfs -text
|
21 |
-
*.pkl filter=lfs diff=lfs merge=lfs -text
|
22 |
-
*.pt filter=lfs diff=lfs merge=lfs -text
|
23 |
-
*.pth filter=lfs diff=lfs merge=lfs -text
|
24 |
-
*.rar filter=lfs diff=lfs merge=lfs -text
|
25 |
-
*.safetensors filter=lfs diff=lfs merge=lfs -text
|
26 |
-
saved_model/**/* filter=lfs diff=lfs merge=lfs -text
|
27 |
-
*.tar.* filter=lfs diff=lfs merge=lfs -text
|
28 |
-
*.tflite filter=lfs diff=lfs merge=lfs -text
|
29 |
-
*.tgz filter=lfs diff=lfs merge=lfs -text
|
30 |
-
*.wasm filter=lfs diff=lfs merge=lfs -text
|
31 |
-
*.xz filter=lfs diff=lfs merge=lfs -text
|
32 |
-
*.zip filter=lfs diff=lfs merge=lfs -text
|
33 |
-
*.zst filter=lfs diff=lfs merge=lfs -text
|
34 |
-
*tfevents* filter=lfs diff=lfs merge=lfs -text
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
README.md
CHANGED
@@ -1,10 +1,10 @@
|
|
1 |
---
|
2 |
title: Inst Inpaint
|
3 |
-
emoji:
|
4 |
colorFrom: purple
|
5 |
colorTo: red
|
6 |
sdk: gradio
|
7 |
-
sdk_version: 3.
|
8 |
app_file: app.py
|
9 |
pinned: false
|
10 |
license: mit
|
|
|
1 |
---
|
2 |
title: Inst Inpaint
|
3 |
+
emoji: 🖌️
|
4 |
colorFrom: purple
|
5 |
colorTo: red
|
6 |
sdk: gradio
|
7 |
+
sdk_version: 3.39.0
|
8 |
app_file: app.py
|
9 |
pinned: false
|
10 |
license: mit
|
app.py
CHANGED
@@ -1,35 +1,53 @@
|
|
|
|
1 |
import gradio as gr
|
2 |
import numpy as np
|
3 |
import torch
|
4 |
from PIL import Image
|
5 |
-
|
6 |
import constants
|
7 |
import utils
|
|
|
|
|
|
|
|
|
|
|
8 |
|
9 |
-
|
10 |
-
|
11 |
|
12 |
-
def inference(image: np.ndarray,
|
13 |
-
|
14 |
-
if not text.lower().startswith("remove the"):
|
15 |
raise gr.Error("Instruction should start with 'Remove the' !")
|
16 |
-
|
17 |
image = Image.fromarray(image)
|
18 |
cropped_image, image = utils.preprocess_image(image, center_crop=center_crop)
|
19 |
-
|
20 |
-
|
21 |
-
prediction = PREDICTOR.predict(image, text, num_steps)
|
22 |
-
|
23 |
-
print("Num steps:", num_steps)
|
24 |
-
|
25 |
-
return cropped_image, prediction
|
26 |
-
|
27 |
|
28 |
if __name__ == "__main__":
|
29 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
30 |
|
31 |
-
|
32 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
33 |
|
34 |
sample_image, sample_instruction, sample_step = constants.EXAMPLES[3]
|
35 |
|
|
|
1 |
+
import argparse
|
2 |
import gradio as gr
|
3 |
import numpy as np
|
4 |
import torch
|
5 |
from PIL import Image
|
|
|
6 |
import constants
|
7 |
import utils
|
8 |
+
from ldm.util import instantiate_from_config
|
9 |
+
from omegaconf import OmegaConf
|
10 |
+
from zipfile import ZipFile
|
11 |
+
import gdown
|
12 |
+
import os
|
13 |
|
14 |
+
MODEL = None
|
|
|
15 |
|
16 |
+
def inference(image: np.ndarray, instruction: str, center_crop: bool):
|
17 |
+
if not instruction.lower().startswith("remove the"):
|
|
|
18 |
raise gr.Error("Instruction should start with 'Remove the' !")
|
|
|
19 |
image = Image.fromarray(image)
|
20 |
cropped_image, image = utils.preprocess_image(image, center_crop=center_crop)
|
21 |
+
output_image = MODEL.inpaint(image, instruction, num_steps=10, device="cpu", return_pil=True, seed=0)
|
22 |
+
return cropped_image, output_image
|
|
|
|
|
|
|
|
|
|
|
|
|
23 |
|
24 |
if __name__ == "__main__":
|
25 |
+
parser = argparse.ArgumentParser()
|
26 |
+
parser.add_argument(
|
27 |
+
"--config",
|
28 |
+
type=str,
|
29 |
+
default="configs/latent-diffusion/gqa-inpaint-ldm-vq-f8-256x256.yaml",
|
30 |
+
help="Path of the model config file",
|
31 |
+
)
|
32 |
+
parser.add_argument(
|
33 |
+
"--checkpoint",
|
34 |
+
type=str,
|
35 |
+
default="models/gqa_inpaint/ldm/model.ckpt",
|
36 |
+
help="Path of the model checkpoint file",
|
37 |
+
)
|
38 |
+
args = parser.parse_args()
|
39 |
|
40 |
+
gdown.download(id="1tp0aHAS-ccrIfNz7XrGTSdNIPNZjOVSp", output="models/")
|
41 |
+
with ZipFile("models/gqa_inpaint.zip", 'r') as zObject:
|
42 |
+
zObject.extractall(path="models/")
|
43 |
+
os.remove("models/gqa_inpaint.zip")
|
44 |
+
|
45 |
+
parsed_config = OmegaConf.load(args.config)
|
46 |
+
MODEL = instantiate_from_config(parsed_config["model"])
|
47 |
+
model_state_dict = torch.load(args.checkpoint, map_location="cpu")["state_dict"]
|
48 |
+
MODEL.load_state_dict(model_state_dict)
|
49 |
+
MODEL.eval()
|
50 |
+
MODEL.to("cpu")
|
51 |
|
52 |
sample_image, sample_instruction, sample_step = constants.EXAMPLES[3]
|
53 |
|
configs/latent-diffusion/gqa-inpaint-ldm-vq-f8-256x256.yaml
ADDED
@@ -0,0 +1,122 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
model:
|
2 |
+
base_learning_rate: 2.0e-06
|
3 |
+
target: ldm.models.diffusion.ddpm.LatentDiffusion
|
4 |
+
params:
|
5 |
+
linear_start: 0.0015
|
6 |
+
linear_end: 0.0195
|
7 |
+
num_timesteps_cond: 1
|
8 |
+
log_every_t: 100
|
9 |
+
timesteps: 1000
|
10 |
+
first_stage_key: "target_image"
|
11 |
+
cond_stage_key: "source_image"
|
12 |
+
cond_stage_trainable: False
|
13 |
+
cond_stage_instruction_key: "text"
|
14 |
+
cond_stage_instruction_embedder_trainable: True
|
15 |
+
conditioning_key: "hybrid"
|
16 |
+
image_size: 32
|
17 |
+
channels: 4
|
18 |
+
monitor: val/loss_simple_ema
|
19 |
+
|
20 |
+
unet_config:
|
21 |
+
target: ldm.modules.diffusionmodules.openaimodel.UNetModel
|
22 |
+
params:
|
23 |
+
image_size: 32
|
24 |
+
in_channels: 8 # 4 (noisy image features) + 4 (source image features)
|
25 |
+
out_channels: 4
|
26 |
+
model_channels: 128
|
27 |
+
attention_resolutions: [8,4,2]
|
28 |
+
num_res_blocks: 2
|
29 |
+
channel_mult: [1,2,3,4]
|
30 |
+
num_heads: 8
|
31 |
+
resblock_updown: True
|
32 |
+
|
33 |
+
###### Instruction embedding cross attention ######
|
34 |
+
use_spatial_transformer: true
|
35 |
+
transformer_depth: 1
|
36 |
+
context_dim: 512
|
37 |
+
###################################################
|
38 |
+
|
39 |
+
first_stage_config:
|
40 |
+
target: ldm.models.autoencoder.VQModel
|
41 |
+
params:
|
42 |
+
ckpt_path: models/gqa_inpaint/first_stage/vq-f8-cb16384-openimages.ckpt
|
43 |
+
monitor: "val/rec_loss"
|
44 |
+
embed_dim: 4
|
45 |
+
n_embed: 16384
|
46 |
+
lossconfig:
|
47 |
+
target: torch.nn.Identity
|
48 |
+
ddconfig:
|
49 |
+
double_z: false
|
50 |
+
z_channels: 4
|
51 |
+
resolution: 256
|
52 |
+
in_channels: 3
|
53 |
+
out_ch: 3
|
54 |
+
ch: 128
|
55 |
+
ch_mult: [1,2,2,4]
|
56 |
+
num_res_blocks: 2
|
57 |
+
attn_resolutions: [32]
|
58 |
+
dropout: 0.0
|
59 |
+
|
60 |
+
cond_stage_config: __is_first_stage__
|
61 |
+
|
62 |
+
cond_stage_instruction_embedder_config:
|
63 |
+
target: ldm.modules.encoders.modules.BERTEmbedder
|
64 |
+
params:
|
65 |
+
n_embed: 512
|
66 |
+
n_layer: 16
|
67 |
+
|
68 |
+
data:
|
69 |
+
target: main.DataModuleFromConfig
|
70 |
+
params:
|
71 |
+
batch_size: 8
|
72 |
+
num_workers: 4
|
73 |
+
|
74 |
+
train:
|
75 |
+
target: dataset.gqa_inpaint.GQAInpaintTrain
|
76 |
+
params:
|
77 |
+
images_root: "data/gqa-inpaint/images"
|
78 |
+
images_inpainted_root: "data/gqa-inpaint/images_inpainted"
|
79 |
+
masks_root: "data/gqa-inpaint/masks"
|
80 |
+
scene_json_path: "data/gqa-inpaint/train_scenes.json"
|
81 |
+
max_relations: 1
|
82 |
+
simplify_augment: True
|
83 |
+
instruction_type: "remove"
|
84 |
+
size: 256
|
85 |
+
irrelevant_text_prob: 0.2
|
86 |
+
|
87 |
+
validation:
|
88 |
+
target: dataset.gqa_inpaint.GQAInpaintTest
|
89 |
+
params:
|
90 |
+
images_root: "data/gqa-inpaint/images"
|
91 |
+
images_inpainted_root: "data/gqa-inpaint/images_inpainted"
|
92 |
+
masks_root: "data/gqa-inpaint/masks"
|
93 |
+
scene_json_path: "data/gqa-inpaint/test_scenes.json"
|
94 |
+
max_relations: 1
|
95 |
+
simplify_augment: True
|
96 |
+
instruction_type: "remove"
|
97 |
+
size: 256
|
98 |
+
|
99 |
+
test:
|
100 |
+
target: dataset.gqa_inpaint.GQAInpaintTest
|
101 |
+
params:
|
102 |
+
images_root: "data/gqa-inpaint/images"
|
103 |
+
images_inpainted_root: "data/gqa-inpaint/images_inpainted"
|
104 |
+
masks_root: "data/gqa-inpaint/masks"
|
105 |
+
scene_json_path: "data/gqa-inpaint/test_scenes.json"
|
106 |
+
test_instructions_path: "data/gqa-inpaint/test_instructions.json"
|
107 |
+
max_relations: 1
|
108 |
+
simplify_augment: True
|
109 |
+
instruction_type: "remove"
|
110 |
+
size: 256
|
111 |
+
|
112 |
+
lightning:
|
113 |
+
callbacks:
|
114 |
+
image_logger:
|
115 |
+
target: main.ImageLogger
|
116 |
+
params:
|
117 |
+
batch_frequency: 5000
|
118 |
+
max_images: 8
|
119 |
+
increase_log_steps: True
|
120 |
+
|
121 |
+
trainer:
|
122 |
+
benchmark: True
|
constants.py
CHANGED
@@ -14,7 +14,7 @@ DESCRIPTION = """
|
|
14 |
EXAMPLES = [
|
15 |
["examples/kite-boy.png", "Remove the colorful kite", True],
|
16 |
["examples/cat-car.jpg", "Remove the car", True],
|
17 |
-
["examples/bus-tree.jpg", "Remove the bus", True],
|
18 |
["examples/cups.webp", "Remove the cup at the left", True],
|
19 |
["examples/woman-fantasy.jpg", "Remove the woman", True],
|
20 |
["examples/clock.png", "Remove the round clock at the center", True],
|
|
|
14 |
EXAMPLES = [
|
15 |
["examples/kite-boy.png", "Remove the colorful kite", True],
|
16 |
["examples/cat-car.jpg", "Remove the car", True],
|
17 |
+
["examples/bus-tree.jpg", "Remove the red bus", True],
|
18 |
["examples/cups.webp", "Remove the cup at the left", True],
|
19 |
["examples/woman-fantasy.jpg", "Remove the woman", True],
|
20 |
["examples/clock.png", "Remove the round clock at the center", True],
|
ldm/__init__.py
ADDED
File without changes
|
ldm/lr_scheduler.py
ADDED
@@ -0,0 +1,98 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import numpy as np
|
2 |
+
|
3 |
+
|
4 |
+
class LambdaWarmUpCosineScheduler:
|
5 |
+
"""
|
6 |
+
note: use with a base_lr of 1.0
|
7 |
+
"""
|
8 |
+
def __init__(self, warm_up_steps, lr_min, lr_max, lr_start, max_decay_steps, verbosity_interval=0):
|
9 |
+
self.lr_warm_up_steps = warm_up_steps
|
10 |
+
self.lr_start = lr_start
|
11 |
+
self.lr_min = lr_min
|
12 |
+
self.lr_max = lr_max
|
13 |
+
self.lr_max_decay_steps = max_decay_steps
|
14 |
+
self.last_lr = 0.
|
15 |
+
self.verbosity_interval = verbosity_interval
|
16 |
+
|
17 |
+
def schedule(self, n, **kwargs):
|
18 |
+
if self.verbosity_interval > 0:
|
19 |
+
if n % self.verbosity_interval == 0: print(f"current step: {n}, recent lr-multiplier: {self.last_lr}")
|
20 |
+
if n < self.lr_warm_up_steps:
|
21 |
+
lr = (self.lr_max - self.lr_start) / self.lr_warm_up_steps * n + self.lr_start
|
22 |
+
self.last_lr = lr
|
23 |
+
return lr
|
24 |
+
else:
|
25 |
+
t = (n - self.lr_warm_up_steps) / (self.lr_max_decay_steps - self.lr_warm_up_steps)
|
26 |
+
t = min(t, 1.0)
|
27 |
+
lr = self.lr_min + 0.5 * (self.lr_max - self.lr_min) * (
|
28 |
+
1 + np.cos(t * np.pi))
|
29 |
+
self.last_lr = lr
|
30 |
+
return lr
|
31 |
+
|
32 |
+
def __call__(self, n, **kwargs):
|
33 |
+
return self.schedule(n,**kwargs)
|
34 |
+
|
35 |
+
|
36 |
+
class LambdaWarmUpCosineScheduler2:
|
37 |
+
"""
|
38 |
+
supports repeated iterations, configurable via lists
|
39 |
+
note: use with a base_lr of 1.0.
|
40 |
+
"""
|
41 |
+
def __init__(self, warm_up_steps, f_min, f_max, f_start, cycle_lengths, verbosity_interval=0):
|
42 |
+
assert len(warm_up_steps) == len(f_min) == len(f_max) == len(f_start) == len(cycle_lengths)
|
43 |
+
self.lr_warm_up_steps = warm_up_steps
|
44 |
+
self.f_start = f_start
|
45 |
+
self.f_min = f_min
|
46 |
+
self.f_max = f_max
|
47 |
+
self.cycle_lengths = cycle_lengths
|
48 |
+
self.cum_cycles = np.cumsum([0] + list(self.cycle_lengths))
|
49 |
+
self.last_f = 0.
|
50 |
+
self.verbosity_interval = verbosity_interval
|
51 |
+
|
52 |
+
def find_in_interval(self, n):
|
53 |
+
interval = 0
|
54 |
+
for cl in self.cum_cycles[1:]:
|
55 |
+
if n <= cl:
|
56 |
+
return interval
|
57 |
+
interval += 1
|
58 |
+
|
59 |
+
def schedule(self, n, **kwargs):
|
60 |
+
cycle = self.find_in_interval(n)
|
61 |
+
n = n - self.cum_cycles[cycle]
|
62 |
+
if self.verbosity_interval > 0:
|
63 |
+
if n % self.verbosity_interval == 0: print(f"current step: {n}, recent lr-multiplier: {self.last_f}, "
|
64 |
+
f"current cycle {cycle}")
|
65 |
+
if n < self.lr_warm_up_steps[cycle]:
|
66 |
+
f = (self.f_max[cycle] - self.f_start[cycle]) / self.lr_warm_up_steps[cycle] * n + self.f_start[cycle]
|
67 |
+
self.last_f = f
|
68 |
+
return f
|
69 |
+
else:
|
70 |
+
t = (n - self.lr_warm_up_steps[cycle]) / (self.cycle_lengths[cycle] - self.lr_warm_up_steps[cycle])
|
71 |
+
t = min(t, 1.0)
|
72 |
+
f = self.f_min[cycle] + 0.5 * (self.f_max[cycle] - self.f_min[cycle]) * (
|
73 |
+
1 + np.cos(t * np.pi))
|
74 |
+
self.last_f = f
|
75 |
+
return f
|
76 |
+
|
77 |
+
def __call__(self, n, **kwargs):
|
78 |
+
return self.schedule(n, **kwargs)
|
79 |
+
|
80 |
+
|
81 |
+
class LambdaLinearScheduler(LambdaWarmUpCosineScheduler2):
|
82 |
+
|
83 |
+
def schedule(self, n, **kwargs):
|
84 |
+
cycle = self.find_in_interval(n)
|
85 |
+
n = n - self.cum_cycles[cycle]
|
86 |
+
if self.verbosity_interval > 0:
|
87 |
+
if n % self.verbosity_interval == 0: print(f"current step: {n}, recent lr-multiplier: {self.last_f}, "
|
88 |
+
f"current cycle {cycle}")
|
89 |
+
|
90 |
+
if n < self.lr_warm_up_steps[cycle]:
|
91 |
+
f = (self.f_max[cycle] - self.f_start[cycle]) / self.lr_warm_up_steps[cycle] * n + self.f_start[cycle]
|
92 |
+
self.last_f = f
|
93 |
+
return f
|
94 |
+
else:
|
95 |
+
f = self.f_min[cycle] + (self.f_max[cycle] - self.f_min[cycle]) * (self.cycle_lengths[cycle] - n) / (self.cycle_lengths[cycle])
|
96 |
+
self.last_f = f
|
97 |
+
return f
|
98 |
+
|
ldm/models/autoencoder.py
ADDED
@@ -0,0 +1,426 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import torch
|
2 |
+
import pytorch_lightning as pl
|
3 |
+
import torch.nn.functional as F
|
4 |
+
from contextlib import contextmanager
|
5 |
+
from packaging import version
|
6 |
+
import numpy as np
|
7 |
+
|
8 |
+
from taming.modules.vqvae.quantize import VectorQuantizer2 as VectorQuantizer
|
9 |
+
|
10 |
+
from ldm.modules.diffusionmodules.model import Encoder, Decoder
|
11 |
+
from ldm.modules.distributions.distributions import DiagonalGaussianDistribution
|
12 |
+
|
13 |
+
from ldm.util import instantiate_from_config
|
14 |
+
|
15 |
+
|
16 |
+
class VQModel(pl.LightningModule):
|
17 |
+
def __init__(self,
|
18 |
+
ddconfig,
|
19 |
+
lossconfig,
|
20 |
+
n_embed,
|
21 |
+
embed_dim,
|
22 |
+
ckpt_path=None,
|
23 |
+
ignore_keys=[],
|
24 |
+
image_key="image",
|
25 |
+
colorize_nlabels=None,
|
26 |
+
monitor=None,
|
27 |
+
batch_resize_range=None,
|
28 |
+
scheduler_config=None,
|
29 |
+
lr_g_factor=1.0,
|
30 |
+
remap=None,
|
31 |
+
sane_index_shape=False, # Telling vector quantizer to return indices
|
32 |
+
use_ema=False
|
33 |
+
):
|
34 |
+
super().__init__()
|
35 |
+
self.embed_dim = embed_dim
|
36 |
+
self.n_embed = n_embed
|
37 |
+
self.image_key = image_key
|
38 |
+
self.encoder = Encoder(**ddconfig)
|
39 |
+
self.decoder = Decoder(**ddconfig)
|
40 |
+
self.loss = instantiate_from_config(lossconfig)
|
41 |
+
self.quantize = VectorQuantizer(n_embed, embed_dim, beta=0.25,
|
42 |
+
remap=remap,
|
43 |
+
sane_index_shape=sane_index_shape)
|
44 |
+
self.quant_conv = torch.nn.Conv2d(ddconfig["z_channels"], embed_dim, 1)
|
45 |
+
self.post_quant_conv = torch.nn.Conv2d(embed_dim, ddconfig["z_channels"], 1)
|
46 |
+
if colorize_nlabels is not None:
|
47 |
+
assert type(colorize_nlabels)==int
|
48 |
+
self.register_buffer("colorize", torch.randn(3, colorize_nlabels, 1, 1))
|
49 |
+
if monitor is not None:
|
50 |
+
self.monitor = monitor
|
51 |
+
self.batch_resize_range = batch_resize_range
|
52 |
+
if self.batch_resize_range is not None:
|
53 |
+
print(f"{self.__class__.__name__}: Using per-batch resizing in range {batch_resize_range}.")
|
54 |
+
|
55 |
+
self.use_ema = use_ema
|
56 |
+
if self.use_ema:
|
57 |
+
self.model_ema = LitEma(self)
|
58 |
+
print(f"Keeping EMAs of {len(list(self.model_ema.buffers()))}.")
|
59 |
+
|
60 |
+
if ckpt_path is not None:
|
61 |
+
self.init_from_ckpt(ckpt_path, ignore_keys=ignore_keys)
|
62 |
+
self.scheduler_config = scheduler_config
|
63 |
+
self.lr_g_factor = lr_g_factor
|
64 |
+
|
65 |
+
@contextmanager
|
66 |
+
def ema_scope(self, context=None):
|
67 |
+
if self.use_ema:
|
68 |
+
self.model_ema.store(self.parameters())
|
69 |
+
self.model_ema.copy_to(self)
|
70 |
+
if context is not None:
|
71 |
+
print(f"{context}: Switched to EMA weights")
|
72 |
+
try:
|
73 |
+
yield None
|
74 |
+
finally:
|
75 |
+
if self.use_ema:
|
76 |
+
self.model_ema.restore(self.parameters())
|
77 |
+
if context is not None:
|
78 |
+
print(f"{context}: Restored training weights")
|
79 |
+
|
80 |
+
def init_from_ckpt(self, path, ignore_keys=list()):
|
81 |
+
sd = torch.load(path, map_location="cpu")["state_dict"]
|
82 |
+
keys = list(sd.keys())
|
83 |
+
for k in keys:
|
84 |
+
for ik in ignore_keys:
|
85 |
+
if k.startswith(ik):
|
86 |
+
print("Deleting key {} from state_dict.".format(k))
|
87 |
+
del sd[k]
|
88 |
+
missing, unexpected = self.load_state_dict(sd, strict=False)
|
89 |
+
print(f"Restored from {path} with {len(missing)} missing and {len(unexpected)} unexpected keys")
|
90 |
+
if len(missing) > 0:
|
91 |
+
print(f"Missing Keys: {missing}")
|
92 |
+
print(f"Unexpected Keys: {unexpected}")
|
93 |
+
|
94 |
+
def on_train_batch_end(self, *args, **kwargs):
|
95 |
+
if self.use_ema:
|
96 |
+
self.model_ema(self)
|
97 |
+
|
98 |
+
def encode(self, x, return_all=False):
|
99 |
+
h = self.encoder(x)
|
100 |
+
h = self.quant_conv(h)
|
101 |
+
quant, emb_loss, info = self.quantize(h)
|
102 |
+
if return_all:
|
103 |
+
return quant, emb_loss, info
|
104 |
+
return quant
|
105 |
+
|
106 |
+
def encode_to_prequant(self, x):
|
107 |
+
h = self.encoder(x)
|
108 |
+
h = self.quant_conv(h)
|
109 |
+
return h
|
110 |
+
|
111 |
+
def decode(self, quant):
|
112 |
+
quant = self.post_quant_conv(quant)
|
113 |
+
dec = self.decoder(quant)
|
114 |
+
return dec
|
115 |
+
|
116 |
+
def decode_code(self, code_b):
|
117 |
+
quant_b = self.quantize.embed_code(code_b)
|
118 |
+
dec = self.decode(quant_b)
|
119 |
+
return dec
|
120 |
+
|
121 |
+
def forward(self, input, return_pred_indices=False):
|
122 |
+
quant, diff, (_,_,ind) = self.encode(input)
|
123 |
+
dec = self.decode(quant)
|
124 |
+
if return_pred_indices:
|
125 |
+
return dec, diff, ind
|
126 |
+
return dec, diff
|
127 |
+
|
128 |
+
def get_input(self, batch, k):
|
129 |
+
x = batch[k]
|
130 |
+
if len(x.shape) == 3:
|
131 |
+
x = x[..., None]
|
132 |
+
x = x.permute(0, 3, 1, 2).to(memory_format=torch.contiguous_format).float()
|
133 |
+
if self.batch_resize_range is not None:
|
134 |
+
lower_size = self.batch_resize_range[0]
|
135 |
+
upper_size = self.batch_resize_range[1]
|
136 |
+
if self.global_step <= 4:
|
137 |
+
new_resize = upper_size
|
138 |
+
else:
|
139 |
+
new_resize = np.random.choice(np.arange(lower_size, upper_size+16, 16))
|
140 |
+
if new_resize != x.shape[2]:
|
141 |
+
x = F.interpolate(x, size=new_resize, mode="bicubic")
|
142 |
+
x = x.detach()
|
143 |
+
return x
|
144 |
+
|
145 |
+
def training_step(self, batch, batch_idx, optimizer_idx):
|
146 |
+
# https://github.com/pytorch/pytorch/issues/37142
|
147 |
+
# Try not to fool the heuristics
|
148 |
+
x = self.get_input(batch, self.image_key)
|
149 |
+
xrec, qloss, ind = self(x, return_pred_indices=True)
|
150 |
+
|
151 |
+
if optimizer_idx == 0:
|
152 |
+
# autoencode
|
153 |
+
aeloss, log_dict_ae = self.loss(qloss, x, xrec, optimizer_idx, self.global_step,
|
154 |
+
last_layer=self.get_last_layer(), split="train",
|
155 |
+
predicted_indices=ind)
|
156 |
+
|
157 |
+
self.log_dict(log_dict_ae, prog_bar=False, logger=True, on_step=True, on_epoch=True)
|
158 |
+
return aeloss
|
159 |
+
|
160 |
+
if optimizer_idx == 1:
|
161 |
+
# Discriminator
|
162 |
+
discloss, log_dict_disc = self.loss(qloss, x, xrec, optimizer_idx, self.global_step,
|
163 |
+
last_layer=self.get_last_layer(), split="train")
|
164 |
+
self.log_dict(log_dict_disc, prog_bar=False, logger=True, on_step=True, on_epoch=True)
|
165 |
+
return discloss
|
166 |
+
|
167 |
+
def validation_step(self, batch, batch_idx):
|
168 |
+
log_dict = self._validation_step(batch, batch_idx)
|
169 |
+
with self.ema_scope():
|
170 |
+
log_dict_ema = self._validation_step(batch, batch_idx, suffix="_ema")
|
171 |
+
return log_dict
|
172 |
+
|
173 |
+
def _validation_step(self, batch, batch_idx, suffix=""):
|
174 |
+
x = self.get_input(batch, self.image_key)
|
175 |
+
xrec, qloss, ind = self(x, return_pred_indices=True)
|
176 |
+
aeloss, log_dict_ae = self.loss(qloss, x, xrec, 0,
|
177 |
+
self.global_step,
|
178 |
+
last_layer=self.get_last_layer(),
|
179 |
+
split="val"+suffix,
|
180 |
+
predicted_indices=ind
|
181 |
+
)
|
182 |
+
|
183 |
+
discloss, log_dict_disc = self.loss(qloss, x, xrec, 1,
|
184 |
+
self.global_step,
|
185 |
+
last_layer=self.get_last_layer(),
|
186 |
+
split="val"+suffix,
|
187 |
+
predicted_indices=ind
|
188 |
+
)
|
189 |
+
rec_loss = log_dict_ae[f"val{suffix}/rec_loss"]
|
190 |
+
self.log(f"val{suffix}/rec_loss", rec_loss,
|
191 |
+
prog_bar=True, logger=True, on_step=False, on_epoch=True, sync_dist=True)
|
192 |
+
self.log(f"val{suffix}/aeloss", aeloss,
|
193 |
+
prog_bar=True, logger=True, on_step=False, on_epoch=True, sync_dist=True)
|
194 |
+
if version.parse(pl.__version__) >= version.parse('1.4.0'):
|
195 |
+
del log_dict_ae[f"val{suffix}/rec_loss"]
|
196 |
+
self.log_dict(log_dict_ae)
|
197 |
+
self.log_dict(log_dict_disc)
|
198 |
+
return self.log_dict
|
199 |
+
|
200 |
+
def configure_optimizers(self):
|
201 |
+
lr_d = self.learning_rate
|
202 |
+
lr_g = self.lr_g_factor*self.learning_rate
|
203 |
+
print("lr_d", lr_d)
|
204 |
+
print("lr_g", lr_g)
|
205 |
+
opt_ae = torch.optim.Adam(list(self.encoder.parameters())+
|
206 |
+
list(self.decoder.parameters())+
|
207 |
+
list(self.quantize.parameters())+
|
208 |
+
list(self.quant_conv.parameters())+
|
209 |
+
list(self.post_quant_conv.parameters()),
|
210 |
+
lr=lr_g, betas=(0.5, 0.9))
|
211 |
+
opt_disc = torch.optim.Adam(self.loss.discriminator.parameters(),
|
212 |
+
lr=lr_d, betas=(0.5, 0.9))
|
213 |
+
|
214 |
+
if self.scheduler_config is not None:
|
215 |
+
scheduler = instantiate_from_config(self.scheduler_config)
|
216 |
+
|
217 |
+
print("Setting up LambdaLR scheduler...")
|
218 |
+
scheduler = [
|
219 |
+
{
|
220 |
+
'scheduler': LambdaLR(opt_ae, lr_lambda=scheduler.schedule),
|
221 |
+
'interval': 'step',
|
222 |
+
'frequency': 1
|
223 |
+
},
|
224 |
+
{
|
225 |
+
'scheduler': LambdaLR(opt_disc, lr_lambda=scheduler.schedule),
|
226 |
+
'interval': 'step',
|
227 |
+
'frequency': 1
|
228 |
+
},
|
229 |
+
]
|
230 |
+
return [opt_ae, opt_disc], scheduler
|
231 |
+
return [opt_ae, opt_disc], []
|
232 |
+
|
233 |
+
def get_last_layer(self):
|
234 |
+
return self.decoder.conv_out.weight
|
235 |
+
|
236 |
+
def log_images(self, batch, only_inputs=False, plot_ema=False, **kwargs):
|
237 |
+
log = dict()
|
238 |
+
x = self.get_input(batch, self.image_key)
|
239 |
+
x = x.to(self.device)
|
240 |
+
if only_inputs:
|
241 |
+
log["inputs"] = x
|
242 |
+
return log
|
243 |
+
xrec, _ = self(x)
|
244 |
+
if x.shape[1] > 3:
|
245 |
+
# Colorize with random projection
|
246 |
+
assert xrec.shape[1] > 3
|
247 |
+
x = self.to_rgb(x)
|
248 |
+
xrec = self.to_rgb(xrec)
|
249 |
+
log["inputs"] = x
|
250 |
+
log["reconstructions"] = xrec
|
251 |
+
if plot_ema:
|
252 |
+
with self.ema_scope():
|
253 |
+
xrec_ema, _ = self(x)
|
254 |
+
if x.shape[1] > 3: xrec_ema = self.to_rgb(xrec_ema)
|
255 |
+
log["reconstructions_ema"] = xrec_ema
|
256 |
+
return log
|
257 |
+
|
258 |
+
def to_rgb(self, x):
|
259 |
+
assert self.image_key == "segmentation"
|
260 |
+
if not hasattr(self, "colorize"):
|
261 |
+
self.register_buffer("colorize", torch.randn(3, x.shape[1], 1, 1).to(x))
|
262 |
+
x = F.conv2d(x, weight=self.colorize)
|
263 |
+
x = 2.*(x-x.min())/(x.max()-x.min()) - 1.
|
264 |
+
return x
|
265 |
+
|
266 |
+
|
267 |
+
class VQModelInterface(VQModel):
|
268 |
+
def __init__(self, embed_dim, *args, **kwargs):
|
269 |
+
super().__init__(embed_dim=embed_dim, *args, **kwargs)
|
270 |
+
self.embed_dim = embed_dim
|
271 |
+
|
272 |
+
def encode(self, x):
|
273 |
+
h = self.encoder(x)
|
274 |
+
h = self.quant_conv(h)
|
275 |
+
return h
|
276 |
+
|
277 |
+
def decode(self, h, force_not_quantize=False):
|
278 |
+
# Also go through quantization layer
|
279 |
+
if not force_not_quantize:
|
280 |
+
quant, emb_loss, info = self.quantize(h)
|
281 |
+
else:
|
282 |
+
quant = h
|
283 |
+
quant = self.post_quant_conv(quant)
|
284 |
+
dec = self.decoder(quant)
|
285 |
+
return dec
|
286 |
+
|
287 |
+
|
288 |
+
class AutoencoderKL(pl.LightningModule):
|
289 |
+
def __init__(self,
|
290 |
+
ddconfig,
|
291 |
+
lossconfig,
|
292 |
+
embed_dim,
|
293 |
+
ckpt_path=None,
|
294 |
+
ignore_keys=[],
|
295 |
+
image_key="image",
|
296 |
+
colorize_nlabels=None,
|
297 |
+
monitor=None,
|
298 |
+
):
|
299 |
+
super().__init__()
|
300 |
+
self.image_key = image_key
|
301 |
+
self.encoder = Encoder(**ddconfig)
|
302 |
+
self.decoder = Decoder(**ddconfig)
|
303 |
+
self.loss = instantiate_from_config(lossconfig)
|
304 |
+
assert ddconfig["double_z"]
|
305 |
+
self.quant_conv = torch.nn.Conv2d(2*ddconfig["z_channels"], 2*embed_dim, 1)
|
306 |
+
self.post_quant_conv = torch.nn.Conv2d(embed_dim, ddconfig["z_channels"], 1)
|
307 |
+
self.embed_dim = embed_dim
|
308 |
+
if colorize_nlabels is not None:
|
309 |
+
assert type(colorize_nlabels)==int
|
310 |
+
self.register_buffer("colorize", torch.randn(3, colorize_nlabels, 1, 1))
|
311 |
+
if monitor is not None:
|
312 |
+
self.monitor = monitor
|
313 |
+
if ckpt_path is not None:
|
314 |
+
self.init_from_ckpt(ckpt_path, ignore_keys=ignore_keys)
|
315 |
+
|
316 |
+
def init_from_ckpt(self, path, ignore_keys=list()):
|
317 |
+
sd = torch.load(path, map_location="cpu")["state_dict"]
|
318 |
+
keys = list(sd.keys())
|
319 |
+
for k in keys:
|
320 |
+
for ik in ignore_keys:
|
321 |
+
if k.startswith(ik):
|
322 |
+
print("Deleting key {} from state_dict.".format(k))
|
323 |
+
del sd[k]
|
324 |
+
self.load_state_dict(sd, strict=False)
|
325 |
+
print(f"Restored from {path}")
|
326 |
+
|
327 |
+
def encode(self, x):
|
328 |
+
h = self.encoder(x)
|
329 |
+
moments = self.quant_conv(h)
|
330 |
+
posterior = DiagonalGaussianDistribution(moments)
|
331 |
+
return posterior
|
332 |
+
|
333 |
+
def decode(self, z):
|
334 |
+
z = self.post_quant_conv(z)
|
335 |
+
dec = self.decoder(z)
|
336 |
+
return dec
|
337 |
+
|
338 |
+
def forward(self, input, sample_posterior=True):
|
339 |
+
posterior = self.encode(input)
|
340 |
+
if sample_posterior:
|
341 |
+
z = posterior.sample()
|
342 |
+
else:
|
343 |
+
z = posterior.mode()
|
344 |
+
dec = self.decode(z)
|
345 |
+
return dec, posterior
|
346 |
+
|
347 |
+
def get_input(self, batch, k):
|
348 |
+
x = batch[k]
|
349 |
+
if len(x.shape) == 3:
|
350 |
+
x = x[..., None]
|
351 |
+
x = x.permute(0, 3, 1, 2).to(memory_format=torch.contiguous_format).float()
|
352 |
+
return x
|
353 |
+
|
354 |
+
def training_step(self, batch, batch_idx, optimizer_idx):
|
355 |
+
inputs = self.get_input(batch, self.image_key)
|
356 |
+
reconstructions, posterior = self(inputs)
|
357 |
+
|
358 |
+
if optimizer_idx == 0:
|
359 |
+
# Training encoder + decoder + logvar
|
360 |
+
aeloss, log_dict_ae = self.loss(inputs, reconstructions, posterior, optimizer_idx, self.global_step,
|
361 |
+
last_layer=self.get_last_layer(), split="train")
|
362 |
+
self.log("aeloss", aeloss, prog_bar=True, logger=True, on_step=True, on_epoch=True)
|
363 |
+
self.log_dict(log_dict_ae, prog_bar=False, logger=True, on_step=True, on_epoch=False)
|
364 |
+
return aeloss
|
365 |
+
|
366 |
+
if optimizer_idx == 1:
|
367 |
+
# Training the discriminator
|
368 |
+
discloss, log_dict_disc = self.loss(inputs, reconstructions, posterior, optimizer_idx, self.global_step,
|
369 |
+
last_layer=self.get_last_layer(), split="train")
|
370 |
+
|
371 |
+
self.log("discloss", discloss, prog_bar=True, logger=True, on_step=True, on_epoch=True)
|
372 |
+
self.log_dict(log_dict_disc, prog_bar=False, logger=True, on_step=True, on_epoch=False)
|
373 |
+
return discloss
|
374 |
+
|
375 |
+
def validation_step(self, batch, batch_idx):
|
376 |
+
inputs = self.get_input(batch, self.image_key)
|
377 |
+
reconstructions, posterior = self(inputs)
|
378 |
+
aeloss, log_dict_ae = self.loss(inputs, reconstructions, posterior, 0, self.global_step,
|
379 |
+
last_layer=self.get_last_layer(), split="val")
|
380 |
+
|
381 |
+
discloss, log_dict_disc = self.loss(inputs, reconstructions, posterior, 1, self.global_step,
|
382 |
+
last_layer=self.get_last_layer(), split="val")
|
383 |
+
|
384 |
+
self.log("val/rec_loss", log_dict_ae["val/rec_loss"])
|
385 |
+
self.log_dict(log_dict_ae)
|
386 |
+
self.log_dict(log_dict_disc)
|
387 |
+
return self.log_dict
|
388 |
+
|
389 |
+
def configure_optimizers(self):
|
390 |
+
lr = self.learning_rate
|
391 |
+
opt_ae = torch.optim.Adam(list(self.encoder.parameters())+
|
392 |
+
list(self.decoder.parameters())+
|
393 |
+
list(self.quant_conv.parameters())+
|
394 |
+
list(self.post_quant_conv.parameters()),
|
395 |
+
lr=lr, betas=(0.5, 0.9))
|
396 |
+
opt_disc = torch.optim.Adam(self.loss.discriminator.parameters(),
|
397 |
+
lr=lr, betas=(0.5, 0.9))
|
398 |
+
return [opt_ae, opt_disc], []
|
399 |
+
|
400 |
+
def get_last_layer(self):
|
401 |
+
return self.decoder.conv_out.weight
|
402 |
+
|
403 |
+
@torch.no_grad()
|
404 |
+
def log_images(self, batch, only_inputs=False, **kwargs):
|
405 |
+
log = dict()
|
406 |
+
x = self.get_input(batch, self.image_key)
|
407 |
+
x = x.to(self.device)
|
408 |
+
if not only_inputs:
|
409 |
+
xrec, posterior = self(x)
|
410 |
+
if x.shape[1] > 3:
|
411 |
+
# Colorize with random projection
|
412 |
+
assert xrec.shape[1] > 3
|
413 |
+
x = self.to_rgb(x)
|
414 |
+
xrec = self.to_rgb(xrec)
|
415 |
+
log["samples"] = self.decode(torch.randn_like(posterior.sample()))
|
416 |
+
log["reconstructions"] = xrec
|
417 |
+
log["inputs"] = x
|
418 |
+
return log
|
419 |
+
|
420 |
+
def to_rgb(self, x):
|
421 |
+
assert self.image_key == "segmentation"
|
422 |
+
if not hasattr(self, "colorize"):
|
423 |
+
self.register_buffer("colorize", torch.randn(3, x.shape[1], 1, 1).to(x))
|
424 |
+
x = F.conv2d(x, weight=self.colorize)
|
425 |
+
x = 2.*(x-x.min())/(x.max()-x.min()) - 1.
|
426 |
+
return x
|
ldm/models/diffusion/ddim.py
ADDED
@@ -0,0 +1,204 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import torch
|
2 |
+
import numpy as np
|
3 |
+
from tqdm import tqdm
|
4 |
+
|
5 |
+
from ldm.modules.diffusionmodules.util import make_ddim_sampling_parameters, make_ddim_timesteps, noise_like
|
6 |
+
|
7 |
+
|
8 |
+
class DDIMSampler(object):
|
9 |
+
def __init__(self, model, schedule="linear", device="cuda", **kwargs):
|
10 |
+
super().__init__()
|
11 |
+
self.model = model
|
12 |
+
self.ddpm_num_timesteps = model.num_timesteps
|
13 |
+
self.schedule = schedule
|
14 |
+
self.device = device
|
15 |
+
|
16 |
+
def register_buffer(self, name, attr):
|
17 |
+
if type(attr) == torch.Tensor:
|
18 |
+
if self.device == "cuda" and attr.device != torch.device("cuda"):
|
19 |
+
attr = attr.to(torch.device("cuda"))
|
20 |
+
setattr(self, name, attr)
|
21 |
+
|
22 |
+
def make_schedule(self, ddim_num_steps, ddim_discretize="uniform", ddim_eta=0., verbose=True):
|
23 |
+
self.ddim_timesteps = make_ddim_timesteps(ddim_discr_method=ddim_discretize, num_ddim_timesteps=ddim_num_steps,
|
24 |
+
num_ddpm_timesteps=self.ddpm_num_timesteps,verbose=verbose)
|
25 |
+
alphas_cumprod = self.model.alphas_cumprod
|
26 |
+
assert alphas_cumprod.shape[0] == self.ddpm_num_timesteps, 'alphas have to be defined for each timestep'
|
27 |
+
to_torch = lambda x: x.clone().detach().to(torch.float32).to(self.model.device)
|
28 |
+
|
29 |
+
self.register_buffer('betas', to_torch(self.model.betas))
|
30 |
+
self.register_buffer('alphas_cumprod', to_torch(alphas_cumprod))
|
31 |
+
self.register_buffer('alphas_cumprod_prev', to_torch(self.model.alphas_cumprod_prev))
|
32 |
+
|
33 |
+
# Calculations for diffusion q(x_t | x_{t-1}) and others
|
34 |
+
self.register_buffer('sqrt_alphas_cumprod', to_torch(np.sqrt(alphas_cumprod.cpu())))
|
35 |
+
self.register_buffer('sqrt_one_minus_alphas_cumprod', to_torch(np.sqrt(1. - alphas_cumprod.cpu())))
|
36 |
+
self.register_buffer('log_one_minus_alphas_cumprod', to_torch(np.log(1. - alphas_cumprod.cpu())))
|
37 |
+
self.register_buffer('sqrt_recip_alphas_cumprod', to_torch(np.sqrt(1. / alphas_cumprod.cpu())))
|
38 |
+
self.register_buffer('sqrt_recipm1_alphas_cumprod', to_torch(np.sqrt(1. / alphas_cumprod.cpu() - 1)))
|
39 |
+
|
40 |
+
# DDIM sampling parameters
|
41 |
+
ddim_sigmas, ddim_alphas, ddim_alphas_prev = make_ddim_sampling_parameters(alphacums=alphas_cumprod.cpu(),
|
42 |
+
ddim_timesteps=self.ddim_timesteps,
|
43 |
+
eta=ddim_eta,verbose=verbose)
|
44 |
+
self.register_buffer('ddim_sigmas', ddim_sigmas)
|
45 |
+
self.register_buffer('ddim_alphas', ddim_alphas)
|
46 |
+
self.register_buffer('ddim_alphas_prev', ddim_alphas_prev)
|
47 |
+
self.register_buffer('ddim_sqrt_one_minus_alphas', np.sqrt(1. - ddim_alphas))
|
48 |
+
sigmas_for_original_sampling_steps = ddim_eta * torch.sqrt(
|
49 |
+
(1 - self.alphas_cumprod_prev) / (1 - self.alphas_cumprod) * (
|
50 |
+
1 - self.alphas_cumprod / self.alphas_cumprod_prev))
|
51 |
+
self.register_buffer('ddim_sigmas_for_original_num_steps', sigmas_for_original_sampling_steps)
|
52 |
+
|
53 |
+
@torch.no_grad()
|
54 |
+
def sample(self,
|
55 |
+
S,
|
56 |
+
batch_size,
|
57 |
+
shape,
|
58 |
+
conditioning=None,
|
59 |
+
callback=None,
|
60 |
+
img_callback=None,
|
61 |
+
quantize_x0=False,
|
62 |
+
eta=0.,
|
63 |
+
mask=None,
|
64 |
+
x0=None,
|
65 |
+
temperature=1.,
|
66 |
+
noise_dropout=0.,
|
67 |
+
score_corrector=None,
|
68 |
+
corrector_kwargs=None,
|
69 |
+
verbose=True,
|
70 |
+
x_T=None,
|
71 |
+
log_every_t=100,
|
72 |
+
unconditional_guidance_scale=1.,
|
73 |
+
unconditional_conditioning=None,
|
74 |
+
keep_attn_maps=False,
|
75 |
+
**kwargs
|
76 |
+
):
|
77 |
+
self.model.keep_attn_map_dict(keep_attn_maps)
|
78 |
+
if conditioning is not None:
|
79 |
+
if isinstance(conditioning, dict):
|
80 |
+
cbs = conditioning[list(conditioning.keys())[0]].shape[0]
|
81 |
+
if cbs != batch_size:
|
82 |
+
print(f"Warning: Got {cbs} conditionings but batch-size is {batch_size}")
|
83 |
+
else:
|
84 |
+
if conditioning.shape[0] != batch_size:
|
85 |
+
print(f"Warning: Got {conditioning.shape[0]} conditionings but batch-size is {batch_size}")
|
86 |
+
|
87 |
+
self.make_schedule(ddim_num_steps=S, ddim_eta=eta, verbose=verbose)
|
88 |
+
# Sampling
|
89 |
+
C, H, W = shape
|
90 |
+
size = (batch_size, C, H, W)
|
91 |
+
print(f'Data shape for DDIM sampling is {size}, eta {eta}')
|
92 |
+
|
93 |
+
samples, intermediates = self.ddim_sampling(conditioning, size,
|
94 |
+
callback=callback,
|
95 |
+
img_callback=img_callback,
|
96 |
+
quantize_denoised=quantize_x0,
|
97 |
+
mask=mask, x0=x0,
|
98 |
+
ddim_use_original_steps=False,
|
99 |
+
noise_dropout=noise_dropout,
|
100 |
+
temperature=temperature,
|
101 |
+
score_corrector=score_corrector,
|
102 |
+
corrector_kwargs=corrector_kwargs,
|
103 |
+
x_T=x_T,
|
104 |
+
log_every_t=log_every_t,
|
105 |
+
unconditional_guidance_scale=unconditional_guidance_scale,
|
106 |
+
unconditional_conditioning=unconditional_conditioning,
|
107 |
+
)
|
108 |
+
return samples, intermediates
|
109 |
+
|
110 |
+
@torch.no_grad()
|
111 |
+
def ddim_sampling(self, cond, shape,
|
112 |
+
x_T=None, ddim_use_original_steps=False,
|
113 |
+
callback=None, timesteps=None, quantize_denoised=False,
|
114 |
+
mask=None, x0=None, img_callback=None, log_every_t=100,
|
115 |
+
temperature=1., noise_dropout=0., score_corrector=None, corrector_kwargs=None,
|
116 |
+
unconditional_guidance_scale=1., unconditional_conditioning=None,):
|
117 |
+
device = self.model.betas.device
|
118 |
+
b = shape[0]
|
119 |
+
if x_T is None:
|
120 |
+
img = torch.randn(shape, device=device)
|
121 |
+
else:
|
122 |
+
img = x_T
|
123 |
+
|
124 |
+
if timesteps is None:
|
125 |
+
timesteps = self.ddpm_num_timesteps if ddim_use_original_steps else self.ddim_timesteps
|
126 |
+
elif timesteps is not None and not ddim_use_original_steps:
|
127 |
+
subset_end = int(min(timesteps / self.ddim_timesteps.shape[0], 1) * self.ddim_timesteps.shape[0]) - 1
|
128 |
+
timesteps = self.ddim_timesteps[:subset_end]
|
129 |
+
|
130 |
+
intermediates = {'x_inter': [img], 'pred_x0': [img]}
|
131 |
+
time_range = reversed(range(0,timesteps)) if ddim_use_original_steps else np.flip(timesteps)
|
132 |
+
total_steps = timesteps if ddim_use_original_steps else timesteps.shape[0]
|
133 |
+
print(f"Running DDIM Sampling with {total_steps} timesteps")
|
134 |
+
|
135 |
+
iterator = tqdm(time_range, desc='DDIM Sampler', total=total_steps)
|
136 |
+
|
137 |
+
for i, step in enumerate(iterator):
|
138 |
+
index = total_steps - i - 1
|
139 |
+
ts = torch.full((b,), step, device=device, dtype=torch.long)
|
140 |
+
|
141 |
+
if mask is not None:
|
142 |
+
assert x0 is not None
|
143 |
+
img_orig = self.model.q_sample(x0, ts)
|
144 |
+
img = img_orig * mask + (1. - mask) * img
|
145 |
+
|
146 |
+
outs = self.p_sample_ddim(img, cond, ts, index=index, use_original_steps=ddim_use_original_steps,
|
147 |
+
quantize_denoised=quantize_denoised, temperature=temperature,
|
148 |
+
noise_dropout=noise_dropout, score_corrector=score_corrector,
|
149 |
+
corrector_kwargs=corrector_kwargs,
|
150 |
+
unconditional_guidance_scale=unconditional_guidance_scale,
|
151 |
+
unconditional_conditioning=unconditional_conditioning)
|
152 |
+
img, pred_x0 = outs
|
153 |
+
if callback: callback(i)
|
154 |
+
if img_callback: img_callback(pred_x0, i)
|
155 |
+
|
156 |
+
if index % log_every_t == 0 or index == total_steps - 1:
|
157 |
+
intermediates['x_inter'].append(img)
|
158 |
+
intermediates['pred_x0'].append(pred_x0)
|
159 |
+
|
160 |
+
return img, intermediates
|
161 |
+
|
162 |
+
@torch.no_grad()
|
163 |
+
def p_sample_ddim(self, x, c, t, index, repeat_noise=False, use_original_steps=False, quantize_denoised=False,
|
164 |
+
temperature=1., noise_dropout=0., score_corrector=None, corrector_kwargs=None,
|
165 |
+
unconditional_guidance_scale=1., unconditional_conditioning=None):
|
166 |
+
b, *_, device = *x.shape, x.device
|
167 |
+
|
168 |
+
if unconditional_conditioning is None or unconditional_guidance_scale == 1.:
|
169 |
+
e_t = self.model.apply_model(x, t, c, index=index)
|
170 |
+
else:
|
171 |
+
x_in = torch.cat([x] * 2)
|
172 |
+
t_in = torch.cat([t] * 2)
|
173 |
+
c_in = torch.cat([unconditional_conditioning, c])
|
174 |
+
e_t_uncond, e_t = self.model.apply_model(x_in, t_in, c_in, index=index).chunk(2)
|
175 |
+
e_t = e_t_uncond + unconditional_guidance_scale * (e_t - e_t_uncond)
|
176 |
+
|
177 |
+
if score_corrector is not None:
|
178 |
+
assert self.model.parameterization == "eps"
|
179 |
+
e_t = score_corrector.modify_score(self.model, e_t, x, t, c, **corrector_kwargs)
|
180 |
+
|
181 |
+
alphas = self.model.alphas_cumprod if use_original_steps else self.ddim_alphas
|
182 |
+
alphas_prev = self.model.alphas_cumprod_prev if use_original_steps else self.ddim_alphas_prev
|
183 |
+
sqrt_one_minus_alphas = self.model.sqrt_one_minus_alphas_cumprod if use_original_steps else self.ddim_sqrt_one_minus_alphas
|
184 |
+
sigmas = self.model.ddim_sigmas_for_original_num_steps if use_original_steps else self.ddim_sigmas
|
185 |
+
|
186 |
+
# Selecting parameters corresponding to the currently considered timestep
|
187 |
+
a_t = torch.full((b, 1, 1, 1), alphas[index], device=device)
|
188 |
+
a_prev = torch.full((b, 1, 1, 1), alphas_prev[index], device=device)
|
189 |
+
sigma_t = torch.full((b, 1, 1, 1), sigmas[index], device=device)
|
190 |
+
sqrt_one_minus_at = torch.full((b, 1, 1, 1), sqrt_one_minus_alphas[index],device=device)
|
191 |
+
|
192 |
+
# Current prediction for x_0
|
193 |
+
pred_x0 = (x - sqrt_one_minus_at * e_t) / a_t.sqrt()
|
194 |
+
if quantize_denoised:
|
195 |
+
pred_x0, _, *_ = self.model.first_stage_model.quantize(pred_x0)
|
196 |
+
|
197 |
+
# Direction pointing to x_t
|
198 |
+
dir_xt = (1. - a_prev - sigma_t**2).sqrt() * e_t
|
199 |
+
noise = sigma_t * noise_like(x.shape, device, repeat_noise) * temperature
|
200 |
+
if noise_dropout > 0.:
|
201 |
+
noise = torch.nn.functional.dropout(noise, p=noise_dropout)
|
202 |
+
x_prev = a_prev.sqrt() * pred_x0 + dir_xt + noise
|
203 |
+
|
204 |
+
return x_prev, pred_x0
|
ldm/models/diffusion/ddpm.py
ADDED
@@ -0,0 +1,1062 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
##################################################################################################
|
2 |
+
# Adapted from: https://github.com/CompVis/latent-diffusion/blob/main/ldm/models/diffusion/ddpm.py
|
3 |
+
##################################################################################################
|
4 |
+
# Utilized resources:
|
5 |
+
# - https://github.com/lucidrains/denoising-diffusion-pytorch/blob/7706bdfc6f527f58d33f84b7b522e61e6e3164b3/denoising_diffusion_pytorch/denoising_diffusion_pytorch.py
|
6 |
+
# - https://github.com/openai/improved-diffusion/blob/e94489283bb876ac1477d5dd7709bbbd2d9902ce/improved_diffusion/gaussian_diffusion.py
|
7 |
+
# - https://github.com/CompVis/taming-transformers
|
8 |
+
##################################################################################################
|
9 |
+
|
10 |
+
import torch
|
11 |
+
import torch.nn as nn
|
12 |
+
import numpy as np
|
13 |
+
import pytorch_lightning as pl
|
14 |
+
from torch.optim.lr_scheduler import LambdaLR
|
15 |
+
from einops import rearrange, repeat
|
16 |
+
from contextlib import contextmanager
|
17 |
+
from functools import partial
|
18 |
+
from tqdm import tqdm
|
19 |
+
from torchvision.utils import make_grid
|
20 |
+
from pytorch_lightning.utilities.distributed import rank_zero_only
|
21 |
+
from ldm.util import log_txt_as_img, exists, default, isimage, mean_flat, count_params, instantiate_from_config
|
22 |
+
from ldm.modules.ema import LitEma
|
23 |
+
from ldm.modules.distributions.distributions import normal_kl, DiagonalGaussianDistribution
|
24 |
+
from ldm.models.autoencoder import VQModelInterface
|
25 |
+
from ldm.modules.diffusionmodules.util import make_beta_schedule, extract_into_tensor, noise_like
|
26 |
+
from ldm.models.diffusion.ddim import DDIMSampler
|
27 |
+
from PIL import Image
|
28 |
+
from ldm.util import seed_everything
|
29 |
+
|
30 |
+
def disabled_train(self, mode=True):
|
31 |
+
"""Overwrite model.train with this function to make sure train/eval mode
|
32 |
+
does not change anymore."""
|
33 |
+
return self
|
34 |
+
|
35 |
+
class DDPM(pl.LightningModule):
|
36 |
+
# DDPM with Gaussian diffusion in image space.
|
37 |
+
def __init__(self,
|
38 |
+
unet_config,
|
39 |
+
timesteps=1000,
|
40 |
+
beta_schedule="linear",
|
41 |
+
loss_type="l2",
|
42 |
+
ckpt_path=None,
|
43 |
+
ignore_keys=[],
|
44 |
+
load_only_unet=False,
|
45 |
+
monitor="val/loss",
|
46 |
+
use_ema=True,
|
47 |
+
first_stage_key="image",
|
48 |
+
image_size=256,
|
49 |
+
channels=3,
|
50 |
+
log_every_t=100,
|
51 |
+
clip_denoised=True,
|
52 |
+
linear_start=1e-4,
|
53 |
+
linear_end=2e-2,
|
54 |
+
cosine_s=8e-3,
|
55 |
+
given_betas=None,
|
56 |
+
original_elbo_weight=0.,
|
57 |
+
v_posterior=0., # Weight for choosing posterior variance as sigma = (1-v) * beta_tilde + v * beta
|
58 |
+
l_simple_weight=1.,
|
59 |
+
conditioning_key=None,
|
60 |
+
parameterization="eps", # All assuming fixed variance schedules
|
61 |
+
scheduler_config=None,
|
62 |
+
learn_logvar=False,
|
63 |
+
logvar_init=0.
|
64 |
+
):
|
65 |
+
super().__init__()
|
66 |
+
assert parameterization in ["eps", "x0"], 'currently only supporting "eps" and "x0"'
|
67 |
+
self.parameterization = parameterization
|
68 |
+
print(f"{self.__class__.__name__}: Running in {self.parameterization}-prediction mode")
|
69 |
+
self.cond_stage_model = None
|
70 |
+
self.clip_denoised = clip_denoised
|
71 |
+
self.log_every_t = log_every_t
|
72 |
+
self.first_stage_key = first_stage_key
|
73 |
+
self.image_size = image_size
|
74 |
+
self.channels = channels
|
75 |
+
self.model = DiffusionWrapper(unet_config, conditioning_key)
|
76 |
+
count_params(self.model, verbose=True)
|
77 |
+
self.use_ema = use_ema
|
78 |
+
if self.use_ema:
|
79 |
+
self.model_ema = LitEma(self.model)
|
80 |
+
print(f"Keeping EMAs of {len(list(self.model_ema.buffers()))}.")
|
81 |
+
|
82 |
+
self.use_scheduler = scheduler_config is not None
|
83 |
+
if self.use_scheduler:
|
84 |
+
self.scheduler_config = scheduler_config
|
85 |
+
|
86 |
+
self.v_posterior = v_posterior
|
87 |
+
self.original_elbo_weight = original_elbo_weight
|
88 |
+
self.l_simple_weight = l_simple_weight
|
89 |
+
|
90 |
+
if monitor is not None:
|
91 |
+
self.monitor = monitor
|
92 |
+
if ckpt_path is not None:
|
93 |
+
self.init_from_ckpt(ckpt_path, ignore_keys=ignore_keys, only_model=load_only_unet)
|
94 |
+
|
95 |
+
self.register_schedule(given_betas=given_betas, beta_schedule=beta_schedule, timesteps=timesteps,
|
96 |
+
linear_start=linear_start, linear_end=linear_end, cosine_s=cosine_s)
|
97 |
+
|
98 |
+
self.loss_type = loss_type
|
99 |
+
|
100 |
+
self.learn_logvar = learn_logvar
|
101 |
+
self.logvar = torch.full(fill_value=logvar_init, size=(self.num_timesteps,))
|
102 |
+
if self.learn_logvar:
|
103 |
+
self.logvar = nn.Parameter(self.logvar, requires_grad=True)
|
104 |
+
|
105 |
+
def register_schedule(self, given_betas=None, beta_schedule="linear", timesteps=1000,
|
106 |
+
linear_start=1e-4, linear_end=2e-2, cosine_s=8e-3):
|
107 |
+
if exists(given_betas):
|
108 |
+
betas = given_betas
|
109 |
+
else:
|
110 |
+
betas = make_beta_schedule(beta_schedule, timesteps, linear_start=linear_start, linear_end=linear_end,
|
111 |
+
cosine_s=cosine_s)
|
112 |
+
alphas = 1. - betas
|
113 |
+
alphas_cumprod = np.cumprod(alphas, axis=0)
|
114 |
+
alphas_cumprod_prev = np.append(1., alphas_cumprod[:-1])
|
115 |
+
|
116 |
+
timesteps, = betas.shape
|
117 |
+
self.num_timesteps = int(timesteps)
|
118 |
+
self.linear_start = linear_start
|
119 |
+
self.linear_end = linear_end
|
120 |
+
assert alphas_cumprod.shape[0] == self.num_timesteps, 'alphas have to be defined for each timestep'
|
121 |
+
|
122 |
+
to_torch = partial(torch.tensor, dtype=torch.float32)
|
123 |
+
|
124 |
+
self.register_buffer('betas', to_torch(betas))
|
125 |
+
self.register_buffer('alphas_cumprod', to_torch(alphas_cumprod))
|
126 |
+
self.register_buffer('alphas_cumprod_prev', to_torch(alphas_cumprod_prev))
|
127 |
+
|
128 |
+
# Calculations for diffusion q(x_t | x_{t-1}) and others
|
129 |
+
self.register_buffer('sqrt_alphas_cumprod', to_torch(np.sqrt(alphas_cumprod)))
|
130 |
+
self.register_buffer('sqrt_one_minus_alphas_cumprod', to_torch(np.sqrt(1. - alphas_cumprod)))
|
131 |
+
self.register_buffer('log_one_minus_alphas_cumprod', to_torch(np.log(1. - alphas_cumprod)))
|
132 |
+
self.register_buffer('sqrt_recip_alphas_cumprod', to_torch(np.sqrt(1. / alphas_cumprod)))
|
133 |
+
self.register_buffer('sqrt_recipm1_alphas_cumprod', to_torch(np.sqrt(1. / alphas_cumprod - 1)))
|
134 |
+
|
135 |
+
# Calculations for posterior q(x_{t-1} | x_t, x_0)
|
136 |
+
# Equal to 1. / (1. / (1. - alpha_cumprod_tm1) + alpha_t / beta_t)
|
137 |
+
posterior_variance = (1 - self.v_posterior) * betas * (1. - alphas_cumprod_prev) / (1. - alphas_cumprod) + self.v_posterior * betas
|
138 |
+
self.register_buffer('posterior_variance', to_torch(posterior_variance))
|
139 |
+
# Log calculation clipped because the posterior variance is 0 at the beginning of the diffusion chain.
|
140 |
+
self.register_buffer('posterior_log_variance_clipped', to_torch(np.log(np.maximum(posterior_variance, 1e-20))))
|
141 |
+
self.register_buffer('posterior_mean_coef1', to_torch(
|
142 |
+
betas * np.sqrt(alphas_cumprod_prev) / (1. - alphas_cumprod)))
|
143 |
+
self.register_buffer('posterior_mean_coef2', to_torch(
|
144 |
+
(1. - alphas_cumprod_prev) * np.sqrt(alphas) / (1. - alphas_cumprod)))
|
145 |
+
|
146 |
+
if self.parameterization == "eps":
|
147 |
+
lvlb_weights = self.betas ** 2 / (
|
148 |
+
2 * self.posterior_variance * to_torch(alphas) * (1 - self.alphas_cumprod))
|
149 |
+
elif self.parameterization == "x0":
|
150 |
+
lvlb_weights = 0.5 * np.sqrt(torch.Tensor(alphas_cumprod)) / (2. * 1 - torch.Tensor(alphas_cumprod))
|
151 |
+
else:
|
152 |
+
raise NotImplementedError("mu not supported")
|
153 |
+
lvlb_weights[0] = lvlb_weights[1]
|
154 |
+
self.register_buffer('lvlb_weights', lvlb_weights, persistent=False)
|
155 |
+
assert not torch.isnan(self.lvlb_weights).all()
|
156 |
+
|
157 |
+
@contextmanager
|
158 |
+
def ema_scope(self, context=None):
|
159 |
+
if self.use_ema:
|
160 |
+
self.model_ema.store(self.model.parameters())
|
161 |
+
self.model_ema.copy_to(self.model)
|
162 |
+
if context is not None:
|
163 |
+
print(f"{context}: Switched to EMA weights")
|
164 |
+
try:
|
165 |
+
yield None
|
166 |
+
finally:
|
167 |
+
if self.use_ema:
|
168 |
+
self.model_ema.restore(self.model.parameters())
|
169 |
+
if context is not None:
|
170 |
+
print(f"{context}: Restored training weights")
|
171 |
+
|
172 |
+
def init_from_ckpt(self, path, ignore_keys=list(), only_model=False):
|
173 |
+
sd = torch.load(path, map_location="cpu")
|
174 |
+
if "state_dict" in list(sd.keys()):
|
175 |
+
sd = sd["state_dict"]
|
176 |
+
keys = list(sd.keys())
|
177 |
+
for k in keys:
|
178 |
+
for ik in ignore_keys:
|
179 |
+
if k.startswith(ik):
|
180 |
+
print("Deleting key {} from state_dict.".format(k))
|
181 |
+
del sd[k]
|
182 |
+
missing, unexpected = self.load_state_dict(sd, strict=False) if not only_model else self.model.load_state_dict(
|
183 |
+
sd, strict=False)
|
184 |
+
print(f"Restored from {path} with {len(missing)} missing and {len(unexpected)} unexpected keys")
|
185 |
+
if len(missing) > 0:
|
186 |
+
print(f"Missing Keys: {missing}")
|
187 |
+
if len(unexpected) > 0:
|
188 |
+
print(f"Unexpected Keys: {unexpected}")
|
189 |
+
|
190 |
+
def q_mean_variance(self, x_start, t):
|
191 |
+
"""
|
192 |
+
Get the distribution q(x_t | x_0).
|
193 |
+
:param x_start: the [N x C x ...] tensor of noiseless inputs.
|
194 |
+
:param t: the number of diffusion steps (minus 1). Here, 0 means one step.
|
195 |
+
:return: A tuple (mean, variance, log_variance), all of x_start's shape.
|
196 |
+
"""
|
197 |
+
mean = (extract_into_tensor(self.sqrt_alphas_cumprod, t, x_start.shape) * x_start)
|
198 |
+
variance = extract_into_tensor(1.0 - self.alphas_cumprod, t, x_start.shape)
|
199 |
+
log_variance = extract_into_tensor(self.log_one_minus_alphas_cumprod, t, x_start.shape)
|
200 |
+
return mean, variance, log_variance
|
201 |
+
|
202 |
+
def predict_start_from_noise(self, x_t, t, noise):
|
203 |
+
return (
|
204 |
+
extract_into_tensor(self.sqrt_recip_alphas_cumprod, t, x_t.shape) * x_t -
|
205 |
+
extract_into_tensor(self.sqrt_recipm1_alphas_cumprod, t, x_t.shape) * noise
|
206 |
+
)
|
207 |
+
|
208 |
+
def q_posterior(self, x_start, x_t, t):
|
209 |
+
posterior_mean = (
|
210 |
+
extract_into_tensor(self.posterior_mean_coef1, t, x_t.shape) * x_start +
|
211 |
+
extract_into_tensor(self.posterior_mean_coef2, t, x_t.shape) * x_t
|
212 |
+
)
|
213 |
+
posterior_variance = extract_into_tensor(self.posterior_variance, t, x_t.shape)
|
214 |
+
posterior_log_variance_clipped = extract_into_tensor(self.posterior_log_variance_clipped, t, x_t.shape)
|
215 |
+
return posterior_mean, posterior_variance, posterior_log_variance_clipped
|
216 |
+
|
217 |
+
def p_mean_variance(self, x, t, clip_denoised: bool):
|
218 |
+
model_out = self.model(x, t)
|
219 |
+
if self.parameterization == "eps":
|
220 |
+
x_recon = self.predict_start_from_noise(x, t=t, noise=model_out)
|
221 |
+
elif self.parameterization == "x0":
|
222 |
+
x_recon = model_out
|
223 |
+
if clip_denoised:
|
224 |
+
x_recon.clamp_(-1., 1.)
|
225 |
+
|
226 |
+
model_mean, posterior_variance, posterior_log_variance = self.q_posterior(x_start=x_recon, x_t=x, t=t)
|
227 |
+
return model_mean, posterior_variance, posterior_log_variance
|
228 |
+
|
229 |
+
@torch.no_grad()
|
230 |
+
def p_sample(self, x, t, clip_denoised=True, repeat_noise=False):
|
231 |
+
b, *_, device = *x.shape, x.device
|
232 |
+
model_mean, _, model_log_variance = self.p_mean_variance(x=x, t=t, clip_denoised=clip_denoised)
|
233 |
+
noise = noise_like(x.shape, device, repeat_noise)
|
234 |
+
# No noise when t == 0
|
235 |
+
nonzero_mask = (1 - (t == 0).float()).reshape(b, *((1,) * (len(x.shape) - 1)))
|
236 |
+
return model_mean + nonzero_mask * (0.5 * model_log_variance).exp() * noise
|
237 |
+
|
238 |
+
@torch.no_grad()
|
239 |
+
def p_sample_loop(self, shape, return_intermediates=False):
|
240 |
+
device = self.betas.device
|
241 |
+
b = shape[0]
|
242 |
+
img = torch.randn(shape, device=device)
|
243 |
+
intermediates = [img]
|
244 |
+
for i in tqdm(reversed(range(0, self.num_timesteps)), desc='Sampling t', total=self.num_timesteps):
|
245 |
+
img = self.p_sample(img, torch.full((b,), i, device=device, dtype=torch.long),
|
246 |
+
clip_denoised=self.clip_denoised)
|
247 |
+
if i % self.log_every_t == 0 or i == self.num_timesteps - 1:
|
248 |
+
intermediates.append(img)
|
249 |
+
if return_intermediates:
|
250 |
+
return img, intermediates
|
251 |
+
return img
|
252 |
+
|
253 |
+
@torch.no_grad()
|
254 |
+
def sample(self, batch_size=16, return_intermediates=False):
|
255 |
+
image_size = self.image_size
|
256 |
+
channels = self.channels
|
257 |
+
return self.p_sample_loop((batch_size, channels, image_size, image_size),
|
258 |
+
return_intermediates=return_intermediates)
|
259 |
+
|
260 |
+
def q_sample(self, x_start, t, noise=None):
|
261 |
+
noise = default(noise, lambda: torch.randn_like(x_start))
|
262 |
+
return (extract_into_tensor(self.sqrt_alphas_cumprod, t, x_start.shape) * x_start +
|
263 |
+
extract_into_tensor(self.sqrt_one_minus_alphas_cumprod, t, x_start.shape) * noise)
|
264 |
+
|
265 |
+
def get_loss(self, pred, target, mean=True):
|
266 |
+
if self.loss_type == 'l1':
|
267 |
+
loss = (target - pred).abs()
|
268 |
+
if mean:
|
269 |
+
loss = loss.mean()
|
270 |
+
elif self.loss_type == 'l2':
|
271 |
+
if mean:
|
272 |
+
loss = torch.nn.functional.mse_loss(target, pred)
|
273 |
+
else:
|
274 |
+
loss = torch.nn.functional.mse_loss(target, pred, reduction='none')
|
275 |
+
else:
|
276 |
+
raise NotImplementedError("unknown loss type '{loss_type}'")
|
277 |
+
|
278 |
+
return loss
|
279 |
+
|
280 |
+
def p_losses(self, x_start, t, noise=None):
|
281 |
+
noise = default(noise, lambda: torch.randn_like(x_start))
|
282 |
+
x_noisy = self.q_sample(x_start=x_start, t=t, noise=noise)
|
283 |
+
model_out = self.model(x_noisy, t)
|
284 |
+
|
285 |
+
loss_dict = {}
|
286 |
+
if self.parameterization == "eps":
|
287 |
+
target = noise
|
288 |
+
elif self.parameterization == "x0":
|
289 |
+
target = x_start
|
290 |
+
else:
|
291 |
+
raise NotImplementedError(f"Paramterization {self.parameterization} not yet supported")
|
292 |
+
|
293 |
+
loss = self.get_loss(model_out, target, mean=False).mean(dim=[1, 2, 3])
|
294 |
+
|
295 |
+
log_prefix = 'train' if self.training else 'val'
|
296 |
+
|
297 |
+
loss_dict.update({f'{log_prefix}/loss_simple': loss.mean()})
|
298 |
+
loss_simple = loss.mean() * self.l_simple_weight
|
299 |
+
|
300 |
+
loss_vlb = (self.lvlb_weights[t] * loss).mean()
|
301 |
+
loss_dict.update({f'{log_prefix}/loss_vlb': loss_vlb})
|
302 |
+
|
303 |
+
loss = loss_simple + self.original_elbo_weight * loss_vlb
|
304 |
+
|
305 |
+
loss_dict.update({f'{log_prefix}/loss': loss})
|
306 |
+
|
307 |
+
return loss, loss_dict
|
308 |
+
|
309 |
+
def forward(self, x, *args, **kwargs):
|
310 |
+
t = torch.randint(0, self.num_timesteps, (x.shape[0],), device=self.device).long()
|
311 |
+
return self.p_losses(x, t, *args, **kwargs)
|
312 |
+
|
313 |
+
def get_input(self, batch, k):
|
314 |
+
x = batch[k]
|
315 |
+
if isinstance(x, list):
|
316 |
+
return x
|
317 |
+
x = x.to(memory_format=torch.contiguous_format).float()
|
318 |
+
return x
|
319 |
+
|
320 |
+
def shared_step(self, batch):
|
321 |
+
x = self.get_input(batch, self.first_stage_key)
|
322 |
+
loss, loss_dict = self(x)
|
323 |
+
return loss, loss_dict
|
324 |
+
|
325 |
+
def training_step(self, batch, batch_idx):
|
326 |
+
loss, loss_dict = self.shared_step(batch)
|
327 |
+
|
328 |
+
self.log_dict(loss_dict, prog_bar=True,
|
329 |
+
logger=True, on_step=True, on_epoch=True)
|
330 |
+
|
331 |
+
self.log("global_step", self.global_step,
|
332 |
+
prog_bar=True, logger=True, on_step=True, on_epoch=False)
|
333 |
+
|
334 |
+
if self.use_scheduler:
|
335 |
+
lr = self.optimizers().param_groups[0]['lr']
|
336 |
+
self.log('lr_abs', lr, prog_bar=True, logger=True, on_step=True, on_epoch=False)
|
337 |
+
|
338 |
+
return loss
|
339 |
+
|
340 |
+
@torch.no_grad()
|
341 |
+
def validation_step(self, batch, batch_idx):
|
342 |
+
_, loss_dict_no_ema = self.shared_step(batch)
|
343 |
+
with self.ema_scope():
|
344 |
+
_, loss_dict_ema = self.shared_step(batch)
|
345 |
+
loss_dict_ema = {key + '_ema': loss_dict_ema[key] for key in loss_dict_ema}
|
346 |
+
self.log_dict(loss_dict_no_ema, prog_bar=False, logger=True, on_step=False, on_epoch=True)
|
347 |
+
self.log_dict(loss_dict_ema, prog_bar=False, logger=True, on_step=False, on_epoch=True)
|
348 |
+
|
349 |
+
def on_train_batch_end(self, *args, **kwargs):
|
350 |
+
if self.use_ema:
|
351 |
+
self.model_ema(self.model)
|
352 |
+
|
353 |
+
def _get_rows_from_list(self, samples):
|
354 |
+
n_imgs_per_row = len(samples)
|
355 |
+
denoise_grid = rearrange(samples, 'n b c h w -> b n c h w')
|
356 |
+
denoise_grid = rearrange(denoise_grid, 'b n c h w -> (b n) c h w')
|
357 |
+
denoise_grid = make_grid(denoise_grid, nrow=n_imgs_per_row)
|
358 |
+
return denoise_grid
|
359 |
+
|
360 |
+
@torch.no_grad()
|
361 |
+
def log_images(self, batch, N=8, n_row=2, sample=True, return_keys=None, **kwargs):
|
362 |
+
log = dict()
|
363 |
+
x = self.get_input(batch, self.first_stage_key)
|
364 |
+
N = min(x.shape[0], N)
|
365 |
+
n_row = min(x.shape[0], n_row)
|
366 |
+
x = x.to(self.device)[:N]
|
367 |
+
log["inputs"] = x
|
368 |
+
|
369 |
+
# Getting diffusion row
|
370 |
+
diffusion_row = list()
|
371 |
+
x_start = x[:n_row]
|
372 |
+
|
373 |
+
for t in range(self.num_timesteps):
|
374 |
+
if t % self.log_every_t == 0 or t == self.num_timesteps - 1:
|
375 |
+
t = repeat(torch.tensor([t]), '1 -> b', b=n_row)
|
376 |
+
t = t.to(self.device).long()
|
377 |
+
noise = torch.randn_like(x_start)
|
378 |
+
x_noisy = self.q_sample(x_start=x_start, t=t, noise=noise)
|
379 |
+
diffusion_row.append(x_noisy)
|
380 |
+
|
381 |
+
log["diffusion_row"] = self._get_rows_from_list(diffusion_row)
|
382 |
+
|
383 |
+
if sample:
|
384 |
+
# Getting denoise row
|
385 |
+
with self.ema_scope("Plotting"):
|
386 |
+
samples, denoise_row = self.sample(batch_size=N, return_intermediates=True)
|
387 |
+
|
388 |
+
log["samples"] = samples
|
389 |
+
log["denoise_row"] = self._get_rows_from_list(denoise_row)
|
390 |
+
|
391 |
+
if return_keys:
|
392 |
+
if np.intersect1d(list(log.keys()), return_keys).shape[0] == 0:
|
393 |
+
return log
|
394 |
+
else:
|
395 |
+
return {key: log[key] for key in return_keys}
|
396 |
+
return log
|
397 |
+
|
398 |
+
def configure_optimizers(self):
|
399 |
+
lr = self.learning_rate
|
400 |
+
params = list(self.model.parameters())
|
401 |
+
if self.learn_logvar:
|
402 |
+
params = params + [self.logvar]
|
403 |
+
opt = torch.optim.AdamW(params, lr=lr)
|
404 |
+
return opt
|
405 |
+
|
406 |
+
|
407 |
+
class LatentDiffusion(DDPM):
|
408 |
+
def __init__(self,
|
409 |
+
first_stage_config,
|
410 |
+
cond_stage_config,
|
411 |
+
cond_stage_instruction_embedder_config=None,
|
412 |
+
num_timesteps_cond=None,
|
413 |
+
cond_stage_key="image",
|
414 |
+
cond_stage_instruction_key=None,
|
415 |
+
cond_stage_trainable=False,
|
416 |
+
cond_stage_instruction_embedder_trainable=False,
|
417 |
+
concat_mode=True,
|
418 |
+
cond_stage_forward=None,
|
419 |
+
conditioning_key=None,
|
420 |
+
scale_factor=1.0,
|
421 |
+
scale_by_std=False,
|
422 |
+
*args, **kwargs):
|
423 |
+
self.num_timesteps_cond = default(num_timesteps_cond, 1)
|
424 |
+
self.scale_by_std = scale_by_std
|
425 |
+
assert self.num_timesteps_cond <= kwargs['timesteps']
|
426 |
+
# For backwards compatibility after implementation of DiffusionWrapper
|
427 |
+
if conditioning_key is None:
|
428 |
+
conditioning_key = 'concat' if concat_mode else 'crossattn'
|
429 |
+
if cond_stage_config == '__is_unconditional__':
|
430 |
+
conditioning_key = None
|
431 |
+
ckpt_path = kwargs.pop("ckpt_path", None)
|
432 |
+
ignore_keys = kwargs.pop("ignore_keys", [])
|
433 |
+
super().__init__(conditioning_key=conditioning_key, *args, **kwargs)
|
434 |
+
self.concat_mode = concat_mode
|
435 |
+
self.cond_stage_trainable = cond_stage_trainable
|
436 |
+
self.cond_stage_key = cond_stage_key
|
437 |
+
self.cond_stage_instruction_key = cond_stage_instruction_key
|
438 |
+
self.cond_stage_instruction_embedder_config = cond_stage_instruction_embedder_config
|
439 |
+
self.cond_stage_instruction_embedder_trainable = cond_stage_instruction_embedder_trainable
|
440 |
+
try:
|
441 |
+
self.num_downs = len(first_stage_config.params.ddconfig.ch_mult) - 1
|
442 |
+
except:
|
443 |
+
self.num_downs = 0
|
444 |
+
if not scale_by_std:
|
445 |
+
self.scale_factor = scale_factor
|
446 |
+
else:
|
447 |
+
self.register_buffer('scale_factor', torch.tensor(scale_factor))
|
448 |
+
self.instantiate_first_stage(first_stage_config)
|
449 |
+
self.instantiate_cond_stage(cond_stage_config)
|
450 |
+
self.instantiate_cond_stage_instruction_embedder(cond_stage_instruction_embedder_config)
|
451 |
+
self.cond_stage_forward = cond_stage_forward
|
452 |
+
self.clip_denoised = False
|
453 |
+
self.bbox_tokenizer = None
|
454 |
+
|
455 |
+
self.restarted_from_ckpt = False
|
456 |
+
if ckpt_path is not None:
|
457 |
+
self.init_from_ckpt(ckpt_path, ignore_keys)
|
458 |
+
self.restarted_from_ckpt = True
|
459 |
+
|
460 |
+
def keep_attn_map_dict(self, keep_attn_maps):
|
461 |
+
self.model.keep_attn_map_dict(keep_attn_maps)
|
462 |
+
|
463 |
+
def get_attn_map_dict(self):
|
464 |
+
return self.model.attn_dict
|
465 |
+
|
466 |
+
def make_cond_schedule(self, ):
|
467 |
+
self.cond_ids = torch.full(size=(self.num_timesteps,), fill_value=self.num_timesteps - 1, dtype=torch.long)
|
468 |
+
ids = torch.round(torch.linspace(0, self.num_timesteps - 1, self.num_timesteps_cond)).long()
|
469 |
+
self.cond_ids[:self.num_timesteps_cond] = ids
|
470 |
+
|
471 |
+
@rank_zero_only
|
472 |
+
@torch.no_grad()
|
473 |
+
def on_train_batch_start(self, batch, batch_idx, dataloader_idx):
|
474 |
+
# Only for the very first batch
|
475 |
+
if self.scale_by_std and self.current_epoch == 0 and self.global_step == 0 and batch_idx == 0 and not self.restarted_from_ckpt:
|
476 |
+
assert self.scale_factor == 1., 'rather not use custom rescaling and std-rescaling simultaneously'
|
477 |
+
# Set rescale weight to 1./std of encodings
|
478 |
+
print("### USING STD-RESCALING ###")
|
479 |
+
x = super().get_input(batch, self.first_stage_key)
|
480 |
+
x = x.to(self.device)
|
481 |
+
encoder_posterior = self.encode_first_stage(x)
|
482 |
+
z = self.get_first_stage_encoding(encoder_posterior).detach()
|
483 |
+
del self.scale_factor
|
484 |
+
self.register_buffer('scale_factor', 1. / z.flatten().std())
|
485 |
+
print(f"setting self.scale_factor to {self.scale_factor}")
|
486 |
+
print("### USING STD-RESCALING ###")
|
487 |
+
|
488 |
+
def register_schedule(self,
|
489 |
+
given_betas=None, beta_schedule="linear", timesteps=1000,
|
490 |
+
linear_start=1e-4, linear_end=2e-2, cosine_s=8e-3):
|
491 |
+
super().register_schedule(given_betas, beta_schedule, timesteps, linear_start, linear_end, cosine_s)
|
492 |
+
|
493 |
+
self.shorten_cond_schedule = self.num_timesteps_cond > 1
|
494 |
+
if self.shorten_cond_schedule:
|
495 |
+
self.make_cond_schedule()
|
496 |
+
|
497 |
+
def instantiate_first_stage(self, config):
|
498 |
+
model = instantiate_from_config(config)
|
499 |
+
self.first_stage_model = model.eval()
|
500 |
+
self.first_stage_model.train = disabled_train
|
501 |
+
for param in self.first_stage_model.parameters():
|
502 |
+
param.requires_grad = False
|
503 |
+
|
504 |
+
def instantiate_cond_stage(self, config):
|
505 |
+
if not self.cond_stage_trainable:
|
506 |
+
if config == "__is_first_stage__":
|
507 |
+
print("Using first stage also as cond stage.")
|
508 |
+
self.cond_stage_model = self.first_stage_model
|
509 |
+
elif config == "__is_unconditional__":
|
510 |
+
print(f"Training {self.__class__.__name__} as an unconditional model.")
|
511 |
+
self.cond_stage_model = None
|
512 |
+
else:
|
513 |
+
model = instantiate_from_config(config)
|
514 |
+
self.cond_stage_model = model.eval()
|
515 |
+
self.cond_stage_model.train = disabled_train
|
516 |
+
for param in self.cond_stage_model.parameters():
|
517 |
+
param.requires_grad = False
|
518 |
+
else:
|
519 |
+
assert config != '__is_first_stage__'
|
520 |
+
assert config != '__is_unconditional__'
|
521 |
+
model = instantiate_from_config(config)
|
522 |
+
self.cond_stage_model = model
|
523 |
+
|
524 |
+
def instantiate_cond_stage_instruction_embedder(self, config):
|
525 |
+
if self.cond_stage_instruction_embedder_config is not None:
|
526 |
+
assert self.cond_stage_instruction_key is not None
|
527 |
+
self.cond_stage_instruction_embedder = instantiate_from_config(config)
|
528 |
+
if not self.cond_stage_instruction_embedder_trainable:
|
529 |
+
self.cond_stage_instruction_embedder = self.cond_stage_instruction_embedder.eval()
|
530 |
+
self.cond_stage_instruction_embedder.train = disabled_train
|
531 |
+
for param in self.cond_stage_instruction_embedder.parameters():
|
532 |
+
param.requires_grad = False
|
533 |
+
|
534 |
+
def _get_denoise_row_from_list(self, samples, desc='', force_no_decoder_quantization=False):
|
535 |
+
denoise_row = []
|
536 |
+
for zd in tqdm(samples, desc=desc):
|
537 |
+
denoise_row.append(self.decode_first_stage(zd.to(self.device),
|
538 |
+
force_not_quantize=force_no_decoder_quantization))
|
539 |
+
n_imgs_per_row = len(denoise_row)
|
540 |
+
denoise_row = torch.stack(denoise_row) # n_log_step, n_row, C, H, W
|
541 |
+
denoise_grid = rearrange(denoise_row, 'n b c h w -> b n c h w')
|
542 |
+
denoise_grid = rearrange(denoise_grid, 'b n c h w -> (b n) c h w')
|
543 |
+
denoise_grid = make_grid(denoise_grid, nrow=n_imgs_per_row)
|
544 |
+
return denoise_grid
|
545 |
+
|
546 |
+
def get_first_stage_encoding(self, encoder_posterior):
|
547 |
+
if isinstance(encoder_posterior, DiagonalGaussianDistribution):
|
548 |
+
z = encoder_posterior.sample()
|
549 |
+
elif isinstance(encoder_posterior, torch.Tensor):
|
550 |
+
z = encoder_posterior
|
551 |
+
else:
|
552 |
+
raise NotImplementedError(f"encoder_posterior of type '{type(encoder_posterior)}' not yet implemented")
|
553 |
+
return self.scale_factor * z
|
554 |
+
|
555 |
+
def get_learned_conditioning(self, c):
|
556 |
+
if self.cond_stage_forward is None:
|
557 |
+
if hasattr(self.cond_stage_model, 'encode') and callable(self.cond_stage_model.encode):
|
558 |
+
c = self.cond_stage_model.encode(c)
|
559 |
+
if isinstance(c, DiagonalGaussianDistribution):
|
560 |
+
c = c.mode()
|
561 |
+
else:
|
562 |
+
c = self.cond_stage_model(c)
|
563 |
+
else:
|
564 |
+
assert hasattr(self.cond_stage_model, self.cond_stage_forward)
|
565 |
+
c = getattr(self.cond_stage_model, self.cond_stage_forward)(c)
|
566 |
+
return c
|
567 |
+
|
568 |
+
@torch.no_grad()
|
569 |
+
def get_main_input(self, batch, k, return_first_stage_outputs, force_c_encode,
|
570 |
+
cond_key, return_original_cond, bs):
|
571 |
+
x = super().get_input(batch, k)
|
572 |
+
check_condition_modification = False
|
573 |
+
if bs is not None:
|
574 |
+
x = x[:bs]
|
575 |
+
x = x.to(self.device)
|
576 |
+
encoder_posterior = self.encode_first_stage(x)
|
577 |
+
z = self.get_first_stage_encoding(encoder_posterior).detach()
|
578 |
+
|
579 |
+
if self.model.conditioning_key is not None:
|
580 |
+
check_condition_modification = True
|
581 |
+
if cond_key is None:
|
582 |
+
cond_key = self.cond_stage_key
|
583 |
+
if cond_key != self.first_stage_key:
|
584 |
+
xc = super().get_input(batch, cond_key).to(self.device)
|
585 |
+
else:
|
586 |
+
xc = x
|
587 |
+
if not self.cond_stage_trainable or force_c_encode:
|
588 |
+
if isinstance(xc, dict) or isinstance(xc, list):
|
589 |
+
c = self.get_learned_conditioning(xc)
|
590 |
+
else:
|
591 |
+
c = self.get_learned_conditioning(xc.to(self.device))
|
592 |
+
else:
|
593 |
+
c = xc
|
594 |
+
if bs is not None:
|
595 |
+
c = c[:bs]
|
596 |
+
else:
|
597 |
+
c = None
|
598 |
+
xc = None
|
599 |
+
out = [z, c]
|
600 |
+
if return_first_stage_outputs:
|
601 |
+
xrec = self.decode_first_stage(z)
|
602 |
+
out.extend([x, xrec])
|
603 |
+
if return_original_cond:
|
604 |
+
out.append(xc)
|
605 |
+
return out, check_condition_modification
|
606 |
+
|
607 |
+
def get_input(self, batch, k, return_first_stage_outputs=False, force_c_encode=False,
|
608 |
+
cond_key=None, return_original_cond=False, bs=None):
|
609 |
+
|
610 |
+
out, check_condition_modification = self.get_main_input(batch, k, return_first_stage_outputs, force_c_encode,
|
611 |
+
cond_key, return_original_cond, bs)
|
612 |
+
c = out[1]
|
613 |
+
# Implemented for inpainting model
|
614 |
+
if check_condition_modification:
|
615 |
+
if self.cond_stage_instruction_key and self.model.conditioning_key == "concat":
|
616 |
+
instructions = super().get_input(batch, self.cond_stage_instruction_key)
|
617 |
+
c = self.cond_stage_instruction_embedder(c, instructions)
|
618 |
+
|
619 |
+
if self.cond_stage_instruction_key and self.model.conditioning_key == "hybrid":
|
620 |
+
instructions = super().get_input(batch, self.cond_stage_instruction_key)
|
621 |
+
# Condition image feature is sent as None to the instruction embedder (instruction embedding is not concatenated)
|
622 |
+
instruction_embedding = self.cond_stage_instruction_embedder(None, instructions)
|
623 |
+
c = {'c_concat': c, 'c_crossattn': instruction_embedding}
|
624 |
+
|
625 |
+
out[1] = c
|
626 |
+
return out
|
627 |
+
|
628 |
+
|
629 |
+
@torch.no_grad()
|
630 |
+
def decode_first_stage(self, z, predict_cids=False, force_not_quantize=False):
|
631 |
+
if predict_cids:
|
632 |
+
if z.dim() == 4:
|
633 |
+
z = torch.argmax(z.exp(), dim=1).long()
|
634 |
+
z = self.first_stage_model.quantize.get_codebook_entry(z, shape=None)
|
635 |
+
z = rearrange(z, 'b h w c -> b c h w').contiguous()
|
636 |
+
|
637 |
+
z = 1. / self.scale_factor * z
|
638 |
+
|
639 |
+
if isinstance(self.first_stage_model, VQModelInterface):
|
640 |
+
return self.first_stage_model.decode(z, force_not_quantize=predict_cids or force_not_quantize)
|
641 |
+
else:
|
642 |
+
return self.first_stage_model.decode(z)
|
643 |
+
|
644 |
+
@torch.no_grad()
|
645 |
+
def encode_first_stage(self, x):
|
646 |
+
return self.first_stage_model.encode(x)
|
647 |
+
|
648 |
+
def shared_step(self, batch, **kwargs):
|
649 |
+
x, c = self.get_input(batch, self.first_stage_key)
|
650 |
+
loss = self(x, c)
|
651 |
+
return loss
|
652 |
+
|
653 |
+
def forward(self, x, c, *args, **kwargs):
|
654 |
+
t = torch.randint(0, self.num_timesteps, (x.shape[0],), device=self.device).long()
|
655 |
+
if self.model.conditioning_key is not None:
|
656 |
+
assert c is not None
|
657 |
+
if self.cond_stage_trainable:
|
658 |
+
c = self.get_learned_conditioning(c)
|
659 |
+
if self.shorten_cond_schedule:
|
660 |
+
tc = self.cond_ids[t].to(self.device)
|
661 |
+
c = self.q_sample(x_start=c, t=tc, noise=torch.randn_like(c.float()))
|
662 |
+
return self.p_losses(x, c, t, *args, **kwargs)
|
663 |
+
|
664 |
+
def apply_model(self, x_noisy, t, cond, index=None):
|
665 |
+
# self.model.conditioning_key is not hybrid
|
666 |
+
if not isinstance(cond, dict):
|
667 |
+
if not isinstance(cond, list):
|
668 |
+
cond = [cond]
|
669 |
+
key = 'c_concat' if self.model.conditioning_key == 'concat' else 'c_crossattn'
|
670 |
+
cond = {key: cond}
|
671 |
+
|
672 |
+
x_recon = self.model(x_noisy, t, **cond, index=index)
|
673 |
+
|
674 |
+
if isinstance(x_recon, tuple):
|
675 |
+
return x_recon[0]
|
676 |
+
else:
|
677 |
+
return x_recon
|
678 |
+
|
679 |
+
def _predict_eps_from_xstart(self, x_t, t, pred_xstart):
|
680 |
+
return (extract_into_tensor(self.sqrt_recip_alphas_cumprod, t, x_t.shape) * x_t - pred_xstart) / \
|
681 |
+
extract_into_tensor(self.sqrt_recipm1_alphas_cumprod, t, x_t.shape)
|
682 |
+
|
683 |
+
def _prior_bpd(self, x_start):
|
684 |
+
"""
|
685 |
+
Get the prior KL term for the variational lower-bound, measured in
|
686 |
+
bits-per-dim.
|
687 |
+
This term can't be optimized, as it only depends on the encoder.
|
688 |
+
:param x_start: the [N x C x ...] tensor of inputs.
|
689 |
+
:return: a batch of [N] KL values (in bits), one per batch element.
|
690 |
+
"""
|
691 |
+
batch_size = x_start.shape[0]
|
692 |
+
t = torch.tensor([self.num_timesteps - 1] * batch_size, device=x_start.device)
|
693 |
+
qt_mean, _, qt_log_variance = self.q_mean_variance(x_start, t)
|
694 |
+
kl_prior = normal_kl(mean1=qt_mean, logvar1=qt_log_variance, mean2=0.0, logvar2=0.0)
|
695 |
+
return mean_flat(kl_prior) / np.log(2.0)
|
696 |
+
|
697 |
+
def p_losses(self, x_start, cond, t, noise=None):
|
698 |
+
noise = default(noise, lambda: torch.randn_like(x_start))
|
699 |
+
x_noisy = self.q_sample(x_start=x_start, t=t, noise=noise)
|
700 |
+
model_output = self.apply_model(x_noisy, t, cond)
|
701 |
+
|
702 |
+
loss_dict = {}
|
703 |
+
prefix = 'train' if self.training else 'val'
|
704 |
+
|
705 |
+
if self.parameterization == "x0":
|
706 |
+
target = x_start
|
707 |
+
elif self.parameterization == "eps":
|
708 |
+
target = noise
|
709 |
+
else:
|
710 |
+
raise NotImplementedError()
|
711 |
+
|
712 |
+
loss_simple = self.get_loss(model_output, target, mean=False).mean([1, 2, 3])
|
713 |
+
loss_dict.update({f'{prefix}/loss_simple': loss_simple.mean()})
|
714 |
+
|
715 |
+
logvar_t = self.logvar[t].to(self.device)
|
716 |
+
|
717 |
+
loss = loss_simple / torch.exp(logvar_t) + logvar_t
|
718 |
+
if self.learn_logvar:
|
719 |
+
loss_dict.update({f'{prefix}/loss_gamma': loss.mean()})
|
720 |
+
loss_dict.update({'logvar': self.logvar.data.mean()})
|
721 |
+
|
722 |
+
loss = self.l_simple_weight * loss.mean()
|
723 |
+
|
724 |
+
loss_vlb = self.get_loss(model_output, target, mean=False).mean(dim=(1, 2, 3))
|
725 |
+
loss_vlb = (self.lvlb_weights[t] * loss_vlb).mean()
|
726 |
+
loss_dict.update({f'{prefix}/loss_vlb': loss_vlb})
|
727 |
+
loss += (self.original_elbo_weight * loss_vlb)
|
728 |
+
loss_dict.update({f'{prefix}/loss': loss})
|
729 |
+
|
730 |
+
return loss, loss_dict
|
731 |
+
|
732 |
+
def p_mean_variance(self, x, c, t, clip_denoised: bool, quantize_denoised=False,
|
733 |
+
return_x0=False, score_corrector=None, corrector_kwargs=None):
|
734 |
+
t_in = t
|
735 |
+
model_out = self.apply_model(x, t_in, c)
|
736 |
+
|
737 |
+
if score_corrector is not None:
|
738 |
+
assert self.parameterization == "eps"
|
739 |
+
model_out = score_corrector.modify_score(self, model_out, x, t, c, **corrector_kwargs)
|
740 |
+
|
741 |
+
|
742 |
+
if self.parameterization == "eps":
|
743 |
+
x_recon = self.predict_start_from_noise(x, t=t, noise=model_out)
|
744 |
+
elif self.parameterization == "x0":
|
745 |
+
x_recon = model_out
|
746 |
+
else:
|
747 |
+
raise NotImplementedError()
|
748 |
+
|
749 |
+
if clip_denoised:
|
750 |
+
x_recon.clamp_(-1., 1.)
|
751 |
+
if quantize_denoised:
|
752 |
+
x_recon, _, [_, _, indices] = self.first_stage_model.quantize(x_recon)
|
753 |
+
model_mean, posterior_variance, posterior_log_variance = self.q_posterior(x_start=x_recon, x_t=x, t=t)
|
754 |
+
|
755 |
+
if return_x0:
|
756 |
+
return model_mean, posterior_variance, posterior_log_variance, x_recon
|
757 |
+
else:
|
758 |
+
return model_mean, posterior_variance, posterior_log_variance
|
759 |
+
|
760 |
+
@torch.no_grad()
|
761 |
+
def p_sample(self, x, c, t, clip_denoised=False, repeat_noise=False, quantize_denoised=False, return_x0=False,
|
762 |
+
temperature=1., noise_dropout=0., score_corrector=None, corrector_kwargs=None):
|
763 |
+
b, *_, device = *x.shape, x.device
|
764 |
+
outputs = self.p_mean_variance(x=x, c=c, t=t, clip_denoised=clip_denoised,
|
765 |
+
quantize_denoised=quantize_denoised,
|
766 |
+
return_x0=return_x0,
|
767 |
+
score_corrector=score_corrector, corrector_kwargs=corrector_kwargs)
|
768 |
+
|
769 |
+
if return_x0:
|
770 |
+
model_mean, _, model_log_variance, x0 = outputs
|
771 |
+
else:
|
772 |
+
model_mean, _, model_log_variance = outputs
|
773 |
+
|
774 |
+
noise = noise_like(x.shape, device, repeat_noise) * temperature
|
775 |
+
if noise_dropout > 0.:
|
776 |
+
noise = torch.nn.functional.dropout(noise, p=noise_dropout)
|
777 |
+
# No noise when t == 0
|
778 |
+
nonzero_mask = (1 - (t == 0).float()).reshape(b, *((1,) * (len(x.shape) - 1)))
|
779 |
+
|
780 |
+
if return_x0:
|
781 |
+
return model_mean + nonzero_mask * (0.5 * model_log_variance).exp() * noise, x0
|
782 |
+
else:
|
783 |
+
return model_mean + nonzero_mask * (0.5 * model_log_variance).exp() * noise
|
784 |
+
|
785 |
+
@torch.no_grad()
|
786 |
+
def progressive_denoising(self, cond, shape, verbose=True, callback=None, quantize_denoised=False,
|
787 |
+
img_callback=None, mask=None, x0=None, temperature=1., noise_dropout=0.,
|
788 |
+
score_corrector=None, corrector_kwargs=None, batch_size=None, x_T=None, start_T=None,
|
789 |
+
log_every_t=None):
|
790 |
+
if not log_every_t:
|
791 |
+
log_every_t = self.log_every_t
|
792 |
+
timesteps = self.num_timesteps
|
793 |
+
if batch_size is not None:
|
794 |
+
b = batch_size if batch_size is not None else shape[0]
|
795 |
+
shape = [batch_size] + list(shape)
|
796 |
+
else:
|
797 |
+
b = batch_size = shape[0]
|
798 |
+
if x_T is None:
|
799 |
+
img = torch.randn(shape, device=self.device)
|
800 |
+
else:
|
801 |
+
img = x_T
|
802 |
+
intermediates = []
|
803 |
+
if cond is not None:
|
804 |
+
if isinstance(cond, dict):
|
805 |
+
cond = {key: cond[key][:batch_size] if not isinstance(cond[key], list) else
|
806 |
+
list(map(lambda x: x[:batch_size], cond[key])) for key in cond}
|
807 |
+
else:
|
808 |
+
cond = [c[:batch_size] for c in cond] if isinstance(cond, list) else cond[:batch_size]
|
809 |
+
|
810 |
+
if start_T is not None:
|
811 |
+
timesteps = min(timesteps, start_T)
|
812 |
+
iterator = tqdm(reversed(range(0, timesteps)), desc='Progressive Generation',
|
813 |
+
total=timesteps) if verbose else reversed(
|
814 |
+
range(0, timesteps))
|
815 |
+
if type(temperature) == float:
|
816 |
+
temperature = [temperature] * timesteps
|
817 |
+
|
818 |
+
for i in iterator:
|
819 |
+
ts = torch.full((b,), i, device=self.device, dtype=torch.long)
|
820 |
+
if self.shorten_cond_schedule:
|
821 |
+
assert self.model.conditioning_key != 'hybrid'
|
822 |
+
tc = self.cond_ids[ts].to(cond.device)
|
823 |
+
cond = self.q_sample(x_start=cond, t=tc, noise=torch.randn_like(cond))
|
824 |
+
|
825 |
+
img, x0_partial = self.p_sample(img, cond, ts,
|
826 |
+
clip_denoised=self.clip_denoised,
|
827 |
+
quantize_denoised=quantize_denoised, return_x0=True,
|
828 |
+
temperature=temperature[i], noise_dropout=noise_dropout,
|
829 |
+
score_corrector=score_corrector, corrector_kwargs=corrector_kwargs)
|
830 |
+
if mask is not None:
|
831 |
+
assert x0 is not None
|
832 |
+
img_orig = self.q_sample(x0, ts)
|
833 |
+
img = img_orig * mask + (1. - mask) * img
|
834 |
+
|
835 |
+
if i % log_every_t == 0 or i == timesteps - 1:
|
836 |
+
intermediates.append(x0_partial)
|
837 |
+
if callback: callback(i)
|
838 |
+
if img_callback: img_callback(img, i)
|
839 |
+
return img, intermediates
|
840 |
+
|
841 |
+
@torch.no_grad()
|
842 |
+
def p_sample_loop(self, cond, shape, return_intermediates=False,
|
843 |
+
x_T=None, verbose=True, callback=None, timesteps=None, quantize_denoised=False,
|
844 |
+
mask=None, x0=None, img_callback=None, start_T=None,
|
845 |
+
log_every_t=None):
|
846 |
+
|
847 |
+
if not log_every_t:
|
848 |
+
log_every_t = self.log_every_t
|
849 |
+
device = self.betas.device
|
850 |
+
b = shape[0]
|
851 |
+
if x_T is None:
|
852 |
+
img = torch.randn(shape, device=device)
|
853 |
+
else:
|
854 |
+
img = x_T
|
855 |
+
|
856 |
+
intermediates = [img]
|
857 |
+
if timesteps is None:
|
858 |
+
timesteps = self.num_timesteps
|
859 |
+
|
860 |
+
if start_T is not None:
|
861 |
+
timesteps = min(timesteps, start_T)
|
862 |
+
iterator = tqdm(reversed(range(0, timesteps)), desc='Sampling t', total=timesteps) if verbose else reversed(
|
863 |
+
range(0, timesteps))
|
864 |
+
|
865 |
+
if mask is not None:
|
866 |
+
assert x0 is not None
|
867 |
+
assert x0.shape[2:3] == mask.shape[2:3] # Spatial size has to match
|
868 |
+
|
869 |
+
for i in iterator:
|
870 |
+
ts = torch.full((b,), i, device=device, dtype=torch.long)
|
871 |
+
if self.shorten_cond_schedule:
|
872 |
+
assert self.model.conditioning_key != 'hybrid'
|
873 |
+
tc = self.cond_ids[ts].to(cond.device)
|
874 |
+
cond = self.q_sample(x_start=cond, t=tc, noise=torch.randn_like(cond))
|
875 |
+
|
876 |
+
img = self.p_sample(img, cond, ts,
|
877 |
+
clip_denoised=self.clip_denoised,
|
878 |
+
quantize_denoised=quantize_denoised)
|
879 |
+
if mask is not None:
|
880 |
+
img_orig = self.q_sample(x0, ts)
|
881 |
+
img = img_orig * mask + (1. - mask) * img
|
882 |
+
|
883 |
+
if i % log_every_t == 0 or i == timesteps - 1:
|
884 |
+
intermediates.append(img)
|
885 |
+
if callback: callback(i)
|
886 |
+
if img_callback: img_callback(img, i)
|
887 |
+
|
888 |
+
if return_intermediates:
|
889 |
+
return img, intermediates
|
890 |
+
return img
|
891 |
+
|
892 |
+
@torch.no_grad()
|
893 |
+
def sample(self, cond, batch_size=16, return_intermediates=False, x_T=None,
|
894 |
+
verbose=True, timesteps=None, quantize_denoised=False,
|
895 |
+
mask=None, x0=None, shape=None,**kwargs):
|
896 |
+
if shape is None:
|
897 |
+
shape = (batch_size, self.channels, self.image_size, self.image_size)
|
898 |
+
if cond is not None:
|
899 |
+
if isinstance(cond, dict):
|
900 |
+
cond = {key: cond[key][:batch_size] if not isinstance(cond[key], list) else
|
901 |
+
list(map(lambda x: x[:batch_size], cond[key])) for key in cond}
|
902 |
+
else:
|
903 |
+
cond = [c[:batch_size] for c in cond] if isinstance(cond, list) else cond[:batch_size]
|
904 |
+
return self.p_sample_loop(cond,
|
905 |
+
shape,
|
906 |
+
return_intermediates=return_intermediates, x_T=x_T,
|
907 |
+
verbose=verbose, timesteps=timesteps, quantize_denoised=quantize_denoised,
|
908 |
+
mask=mask, x0=x0)
|
909 |
+
|
910 |
+
@torch.no_grad()
|
911 |
+
def log_images(self, batch, N=8, n_row=4, plot_progressive_rows=True, instruction_img_size=256, **kwargs):
|
912 |
+
|
913 |
+
log = dict()
|
914 |
+
z, c, x, xrec, xc = self.get_input(batch, self.first_stage_key,
|
915 |
+
return_first_stage_outputs=True,
|
916 |
+
force_c_encode=True,
|
917 |
+
return_original_cond=True,
|
918 |
+
bs=N)
|
919 |
+
N = min(x.shape[0], N)
|
920 |
+
n_row = min(x.shape[0], n_row)
|
921 |
+
log["inputs"] = x
|
922 |
+
log["reconstruction"] = xrec
|
923 |
+
if self.model.conditioning_key is not None:
|
924 |
+
if hasattr(self.cond_stage_model, "decode"):
|
925 |
+
if self.cond_stage_instruction_key and self.model.conditioning_key == "concat":
|
926 |
+
c_cond = c[:,:-self.cond_stage_instruction_embedder.out_size,:,:]
|
927 |
+
else:
|
928 |
+
c_cond = c
|
929 |
+
if isinstance(c_cond, dict):
|
930 |
+
c_cond = c_cond["c_concat"]
|
931 |
+
xc = self.cond_stage_model.decode(c_cond)
|
932 |
+
log["conditioning"] = xc
|
933 |
+
elif isimage(xc):
|
934 |
+
log["conditioning"] = xc
|
935 |
+
|
936 |
+
if self.cond_stage_instruction_key is not None:
|
937 |
+
instructions = super().get_input(batch, self.cond_stage_instruction_key)
|
938 |
+
instructions_img = log_txt_as_img((instruction_img_size, instruction_img_size), instructions)
|
939 |
+
log['instructions'] = instructions_img
|
940 |
+
|
941 |
+
if plot_progressive_rows:
|
942 |
+
with self.ema_scope("Plotting Progressives"):
|
943 |
+
img, progressives = self.progressive_denoising(c,
|
944 |
+
shape=(self.channels, self.image_size, self.image_size),
|
945 |
+
batch_size=N)
|
946 |
+
prog_row = self._get_denoise_row_from_list(progressives, desc="Progressive Generation")
|
947 |
+
log["progressive_row"] = prog_row
|
948 |
+
|
949 |
+
return log
|
950 |
+
|
951 |
+
@torch.no_grad()
|
952 |
+
def inpaint(self, image, instruction, num_steps=50, device="cuda", return_pil=True, seed=0):
|
953 |
+
assert len(image.shape) == 4 and image.shape[0] == 1, "Input image should be a tensor object with batch size 1"
|
954 |
+
assert isinstance(instruction, str), "Input instruction type should be String"
|
955 |
+
assert self.model.conditioning_key == "hybrid", "Inpaint function is only available for hybrid conditioning"
|
956 |
+
|
957 |
+
image = image.to(device)
|
958 |
+
sampler = DDIMSampler(self, device=device)
|
959 |
+
|
960 |
+
seed_everything(seed)
|
961 |
+
with torch.no_grad():
|
962 |
+
with self.ema_scope():
|
963 |
+
c = self.get_first_stage_encoding(self.cond_stage_model.encode(image))
|
964 |
+
shape = c.shape[1:]
|
965 |
+
instruction_embedding = self.cond_stage_instruction_embedder(None, [instruction])
|
966 |
+
c = {'c_concat': c, 'c_crossattn': instruction_embedding}
|
967 |
+
batch_size=c["c_concat"].shape[0]
|
968 |
+
output_latent, _ = sampler.sample(S=num_steps,
|
969 |
+
conditioning=c,
|
970 |
+
batch_size=batch_size,
|
971 |
+
shape=shape,
|
972 |
+
verbose=False)
|
973 |
+
output_image_tensor = self.decode_first_stage(output_latent)[0]
|
974 |
+
output_image_tensor = torch.clip(output_image_tensor, -1, 1)
|
975 |
+
output_image_np = ((output_image_tensor + 1) * 127.5).cpu().numpy()
|
976 |
+
output_image = Image.fromarray(output_image_np.transpose(1,2,0).astype(np.uint8))
|
977 |
+
|
978 |
+
if return_pil:
|
979 |
+
return output_image
|
980 |
+
return output_image_tensor
|
981 |
+
|
982 |
+
def configure_optimizers(self):
|
983 |
+
lr = self.learning_rate
|
984 |
+
params = list(self.model.parameters())
|
985 |
+
if self.cond_stage_trainable:
|
986 |
+
print(f"{self.__class__.__name__}: Also optimizing conditioner params!")
|
987 |
+
params = params + list(self.cond_stage_model.parameters())
|
988 |
+
if self.cond_stage_instruction_embedder_trainable:
|
989 |
+
print(f"{self.__class__.__name__}: Also optimizing conditionaer (instruction embedder) params!")
|
990 |
+
params = params + list(self.cond_stage_instruction_embedder.parameters())
|
991 |
+
if self.learn_logvar:
|
992 |
+
print('Diffusion model optimizing logvar')
|
993 |
+
params.append(self.logvar)
|
994 |
+
opt = torch.optim.AdamW(params, lr=lr)
|
995 |
+
if self.use_scheduler:
|
996 |
+
assert 'target' in self.scheduler_config
|
997 |
+
scheduler = instantiate_from_config(self.scheduler_config)
|
998 |
+
|
999 |
+
print("Setting up LambdaLR scheduler...")
|
1000 |
+
scheduler = [
|
1001 |
+
{
|
1002 |
+
'scheduler': LambdaLR(opt, lr_lambda=scheduler.schedule),
|
1003 |
+
'interval': 'step',
|
1004 |
+
'frequency': 1
|
1005 |
+
}]
|
1006 |
+
return [opt], scheduler
|
1007 |
+
return opt
|
1008 |
+
|
1009 |
+
|
1010 |
+
class DiffusionWrapper(pl.LightningModule):
|
1011 |
+
def __init__(self, diff_model_config, conditioning_key):
|
1012 |
+
super().__init__()
|
1013 |
+
self.diffusion_model = instantiate_from_config(diff_model_config)
|
1014 |
+
self.conditioning_key = conditioning_key
|
1015 |
+
assert self.conditioning_key in [None, 'concat', 'crossattn', 'hybrid', 'adm']
|
1016 |
+
self.attn_dict = None
|
1017 |
+
self.keep_attn_maps = False
|
1018 |
+
|
1019 |
+
def keep_attn_map_dict(self, keep_attn_maps):
|
1020 |
+
self.keep_attn_maps = keep_attn_maps
|
1021 |
+
if keep_attn_maps:
|
1022 |
+
if self.attn_dict is None:
|
1023 |
+
self.attn_dict = {}
|
1024 |
+
else:
|
1025 |
+
self.attn_dict.clear()
|
1026 |
+
else:
|
1027 |
+
self.attn_dict = None
|
1028 |
+
|
1029 |
+
def forward(self, x, t, c_concat: list = None, c_crossattn: list = None, c_mask: list = None, index=None):
|
1030 |
+
if self.keep_attn_maps:
|
1031 |
+
assert index is not None
|
1032 |
+
if index not in self.attn_dict:
|
1033 |
+
self.attn_dict[index] = {}
|
1034 |
+
else:
|
1035 |
+
raise Exception("Attention maps of the current time index has already been assigned.")
|
1036 |
+
if self.conditioning_key is None:
|
1037 |
+
out = self.diffusion_model(x, t)
|
1038 |
+
elif self.conditioning_key == 'concat':
|
1039 |
+
if not isinstance(c_concat, list):
|
1040 |
+
c_concat = [c_concat]
|
1041 |
+
xc = torch.cat([x] + c_concat, dim=1)
|
1042 |
+
out = self.diffusion_model(xc, t)
|
1043 |
+
elif self.conditioning_key == 'crossattn':
|
1044 |
+
cc = torch.cat(c_crossattn, 1)
|
1045 |
+
if self.keep_attn_maps:
|
1046 |
+
out = self.diffusion_model(x, t, context=cc, attn_dict=self.attn_dict[index])
|
1047 |
+
else:
|
1048 |
+
out = self.diffusion_model(x, t, context=cc)
|
1049 |
+
elif self.conditioning_key == 'hybrid':
|
1050 |
+
if not isinstance(c_concat, list):
|
1051 |
+
c_concat = [c_concat]
|
1052 |
+
if not isinstance(c_crossattn, list):
|
1053 |
+
c_crossattn = [c_crossattn]
|
1054 |
+
xc = torch.cat([x] + c_concat, dim=1)
|
1055 |
+
cc = torch.cat(c_crossattn, 1)
|
1056 |
+
if self.keep_attn_maps:
|
1057 |
+
out = self.diffusion_model(xc, t, context=cc, attn_dict=self.attn_dict[index])
|
1058 |
+
else:
|
1059 |
+
out = self.diffusion_model(xc, t, context=cc)
|
1060 |
+
else:
|
1061 |
+
raise NotImplementedError()
|
1062 |
+
return out
|
ldm/modules/attention.py
ADDED
@@ -0,0 +1,271 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from inspect import isfunction
|
2 |
+
import math
|
3 |
+
import torch
|
4 |
+
import torch.nn.functional as F
|
5 |
+
from torch import nn, einsum
|
6 |
+
from einops import rearrange, repeat
|
7 |
+
|
8 |
+
from ldm.modules.diffusionmodules.util import checkpoint
|
9 |
+
|
10 |
+
|
11 |
+
def exists(val):
|
12 |
+
return val is not None
|
13 |
+
|
14 |
+
|
15 |
+
def uniq(arr):
|
16 |
+
return{el: True for el in arr}.keys()
|
17 |
+
|
18 |
+
|
19 |
+
def default(val, d):
|
20 |
+
if exists(val):
|
21 |
+
return val
|
22 |
+
return d() if isfunction(d) else d
|
23 |
+
|
24 |
+
|
25 |
+
def max_neg_value(t):
|
26 |
+
return -torch.finfo(t.dtype).max
|
27 |
+
|
28 |
+
|
29 |
+
def init_(tensor):
|
30 |
+
dim = tensor.shape[-1]
|
31 |
+
std = 1 / math.sqrt(dim)
|
32 |
+
tensor.uniform_(-std, std)
|
33 |
+
return tensor
|
34 |
+
|
35 |
+
|
36 |
+
# feedforward
|
37 |
+
class GEGLU(nn.Module):
|
38 |
+
def __init__(self, dim_in, dim_out):
|
39 |
+
super().__init__()
|
40 |
+
self.proj = nn.Linear(dim_in, dim_out * 2)
|
41 |
+
|
42 |
+
def forward(self, x):
|
43 |
+
x, gate = self.proj(x).chunk(2, dim=-1)
|
44 |
+
return x * F.gelu(gate)
|
45 |
+
|
46 |
+
|
47 |
+
class FeedForward(nn.Module):
|
48 |
+
def __init__(self, dim, dim_out=None, mult=4, glu=False, dropout=0.):
|
49 |
+
super().__init__()
|
50 |
+
inner_dim = int(dim * mult)
|
51 |
+
dim_out = default(dim_out, dim)
|
52 |
+
project_in = nn.Sequential(
|
53 |
+
nn.Linear(dim, inner_dim),
|
54 |
+
nn.GELU()
|
55 |
+
) if not glu else GEGLU(dim, inner_dim)
|
56 |
+
|
57 |
+
self.net = nn.Sequential(
|
58 |
+
project_in,
|
59 |
+
nn.Dropout(dropout),
|
60 |
+
nn.Linear(inner_dim, dim_out)
|
61 |
+
)
|
62 |
+
|
63 |
+
def forward(self, x):
|
64 |
+
return self.net(x)
|
65 |
+
|
66 |
+
|
67 |
+
def zero_module(module):
|
68 |
+
"""
|
69 |
+
Zero out the parameters of a module and return it.
|
70 |
+
"""
|
71 |
+
for p in module.parameters():
|
72 |
+
p.detach().zero_()
|
73 |
+
return module
|
74 |
+
|
75 |
+
|
76 |
+
def Normalize(in_channels):
|
77 |
+
return torch.nn.GroupNorm(num_groups=32, num_channels=in_channels, eps=1e-6, affine=True)
|
78 |
+
|
79 |
+
|
80 |
+
class LinearAttention(nn.Module):
|
81 |
+
def __init__(self, dim, heads=4, dim_head=32):
|
82 |
+
super().__init__()
|
83 |
+
self.heads = heads
|
84 |
+
hidden_dim = dim_head * heads
|
85 |
+
self.to_qkv = nn.Conv2d(dim, hidden_dim * 3, 1, bias = False)
|
86 |
+
self.to_out = nn.Conv2d(hidden_dim, dim, 1)
|
87 |
+
|
88 |
+
def forward(self, x):
|
89 |
+
b, c, h, w = x.shape
|
90 |
+
qkv = self.to_qkv(x)
|
91 |
+
q, k, v = rearrange(qkv, 'b (qkv heads c) h w -> qkv b heads c (h w)', heads = self.heads, qkv=3)
|
92 |
+
k = k.softmax(dim=-1)
|
93 |
+
context = torch.einsum('bhdn,bhen->bhde', k, v)
|
94 |
+
out = torch.einsum('bhde,bhdn->bhen', context, q)
|
95 |
+
out = rearrange(out, 'b heads c (h w) -> b (heads c) h w', heads=self.heads, h=h, w=w)
|
96 |
+
return self.to_out(out)
|
97 |
+
|
98 |
+
|
99 |
+
class SpatialSelfAttention(nn.Module):
|
100 |
+
def __init__(self, in_channels):
|
101 |
+
super().__init__()
|
102 |
+
self.in_channels = in_channels
|
103 |
+
|
104 |
+
self.norm = Normalize(in_channels)
|
105 |
+
self.q = torch.nn.Conv2d(in_channels,
|
106 |
+
in_channels,
|
107 |
+
kernel_size=1,
|
108 |
+
stride=1,
|
109 |
+
padding=0)
|
110 |
+
self.k = torch.nn.Conv2d(in_channels,
|
111 |
+
in_channels,
|
112 |
+
kernel_size=1,
|
113 |
+
stride=1,
|
114 |
+
padding=0)
|
115 |
+
self.v = torch.nn.Conv2d(in_channels,
|
116 |
+
in_channels,
|
117 |
+
kernel_size=1,
|
118 |
+
stride=1,
|
119 |
+
padding=0)
|
120 |
+
self.proj_out = torch.nn.Conv2d(in_channels,
|
121 |
+
in_channels,
|
122 |
+
kernel_size=1,
|
123 |
+
stride=1,
|
124 |
+
padding=0)
|
125 |
+
|
126 |
+
def forward(self, x):
|
127 |
+
h_ = x
|
128 |
+
h_ = self.norm(h_)
|
129 |
+
q = self.q(h_)
|
130 |
+
k = self.k(h_)
|
131 |
+
v = self.v(h_)
|
132 |
+
|
133 |
+
# compute attention
|
134 |
+
b,c,h,w = q.shape
|
135 |
+
q = rearrange(q, 'b c h w -> b (h w) c')
|
136 |
+
k = rearrange(k, 'b c h w -> b c (h w)')
|
137 |
+
w_ = torch.einsum('bij,bjk->bik', q, k)
|
138 |
+
|
139 |
+
w_ = w_ * (int(c)**(-0.5))
|
140 |
+
w_ = torch.nn.functional.softmax(w_, dim=2)
|
141 |
+
|
142 |
+
# attend to values
|
143 |
+
v = rearrange(v, 'b c h w -> b c (h w)')
|
144 |
+
w_ = rearrange(w_, 'b i j -> b j i')
|
145 |
+
h_ = torch.einsum('bij,bjk->bik', v, w_)
|
146 |
+
h_ = rearrange(h_, 'b c (h w) -> b c h w', h=h)
|
147 |
+
h_ = self.proj_out(h_)
|
148 |
+
|
149 |
+
return x+h_
|
150 |
+
|
151 |
+
|
152 |
+
class CrossAttention(nn.Module):
|
153 |
+
def __init__(self, query_dim, context_dim=None, heads=8, dim_head=64, dropout=0.):
|
154 |
+
super().__init__()
|
155 |
+
inner_dim = dim_head * heads
|
156 |
+
context_dim = default(context_dim, query_dim)
|
157 |
+
|
158 |
+
self.scale = dim_head ** -0.5
|
159 |
+
self.heads = heads
|
160 |
+
|
161 |
+
self.to_q = nn.Linear(query_dim, inner_dim, bias=False)
|
162 |
+
self.to_k = nn.Linear(context_dim, inner_dim, bias=False)
|
163 |
+
self.to_v = nn.Linear(context_dim, inner_dim, bias=False)
|
164 |
+
|
165 |
+
self.to_out = nn.Sequential(
|
166 |
+
nn.Linear(inner_dim, query_dim),
|
167 |
+
nn.Dropout(dropout)
|
168 |
+
)
|
169 |
+
|
170 |
+
def __add_attention_to_dict(self, attn_dict, layer_type, attn):
|
171 |
+
layer_key = "{}_layer".format(layer_type)
|
172 |
+
if layer_key not in attn_dict:
|
173 |
+
attn_dict[layer_key] = []
|
174 |
+
attn_dict[layer_key].append(attn.cpu())
|
175 |
+
|
176 |
+
def forward(self, x, context=None, mask=None, attn_dict=None, layer_type=None):
|
177 |
+
# Dimensions of the Simge dataset are written in comments
|
178 |
+
h = self.heads
|
179 |
+
q = self.to_q(x)
|
180 |
+
context = default(context, x)
|
181 |
+
|
182 |
+
# NOTE: Rest of the dimensions are reported for the cross-attn case
|
183 |
+
|
184 |
+
k = self.to_k(context)
|
185 |
+
v = self.to_v(context)
|
186 |
+
q, k, v = map(lambda t: rearrange(t, 'b n (h d) -> (b h) n d', h=h), (q, k, v))
|
187 |
+
sim = einsum('b i d, b j d -> b i j', q, k) * self.scale
|
188 |
+
|
189 |
+
if exists(mask):
|
190 |
+
mask = rearrange(mask, 'b ... -> b (...)')
|
191 |
+
max_neg_value = -torch.finfo(sim.dtype).max
|
192 |
+
mask = repeat(mask, 'b j -> (b h) () j', h=h)
|
193 |
+
sim.masked_fill_(~mask, max_neg_value)
|
194 |
+
|
195 |
+
# attention, what we cannot get enough of
|
196 |
+
attn = sim.softmax(dim=-1)
|
197 |
+
|
198 |
+
if context is not None and attn_dict is not None and layer_type is not None:
|
199 |
+
self.__add_attention_to_dict(attn_dict, layer_type, attn)
|
200 |
+
|
201 |
+
out = einsum('b i j, b j d -> b i d', attn, v)
|
202 |
+
out = rearrange(out, '(b h) n d -> b n (h d)', h=h)
|
203 |
+
return self.to_out(out)
|
204 |
+
|
205 |
+
|
206 |
+
class BasicTransformerBlock(nn.Module):
|
207 |
+
def __init__(self, dim, n_heads, d_head, dropout=0., context_dim=None, gated_ff=True, checkpoint=True):
|
208 |
+
super().__init__()
|
209 |
+
self.attn1 = CrossAttention(query_dim=dim, heads=n_heads, dim_head=d_head, dropout=dropout) # is a self-attention
|
210 |
+
self.ff = FeedForward(dim, dropout=dropout, glu=gated_ff)
|
211 |
+
self.attn2 = CrossAttention(query_dim=dim, context_dim=context_dim,
|
212 |
+
heads=n_heads, dim_head=d_head, dropout=dropout) # is self-attn if context is none
|
213 |
+
self.norm1 = nn.LayerNorm(dim)
|
214 |
+
self.norm2 = nn.LayerNorm(dim)
|
215 |
+
self.norm3 = nn.LayerNorm(dim)
|
216 |
+
self.checkpoint = checkpoint
|
217 |
+
|
218 |
+
def forward(self, x, context=None, attn_dict=None, layer_type=None):
|
219 |
+
return checkpoint(self._forward, (x, context, attn_dict, layer_type), self.parameters(), self.checkpoint)
|
220 |
+
|
221 |
+
def _forward(self, x, context=None, attn_dict=None, layer_type=None):
|
222 |
+
x = self.attn1(self.norm1(x)) + x
|
223 |
+
x = self.attn2(self.norm2(x), context=context, attn_dict=attn_dict, layer_type=layer_type) + x
|
224 |
+
x = self.ff(self.norm3(x)) + x
|
225 |
+
return x
|
226 |
+
|
227 |
+
|
228 |
+
class SpatialTransformer(nn.Module):
|
229 |
+
"""
|
230 |
+
Transformer block for image-like data.
|
231 |
+
First, project the input (aka embedding)
|
232 |
+
and reshape to b, t, d.
|
233 |
+
Then apply standard transformer action.
|
234 |
+
Finally, reshape to image
|
235 |
+
"""
|
236 |
+
def __init__(self, in_channels, n_heads, d_head,
|
237 |
+
depth=1, dropout=0., context_dim=None):
|
238 |
+
super().__init__()
|
239 |
+
self.in_channels = in_channels
|
240 |
+
inner_dim = n_heads * d_head
|
241 |
+
self.norm = Normalize(in_channels)
|
242 |
+
|
243 |
+
self.proj_in = nn.Conv2d(in_channels,
|
244 |
+
inner_dim,
|
245 |
+
kernel_size=1,
|
246 |
+
stride=1,
|
247 |
+
padding=0)
|
248 |
+
|
249 |
+
self.transformer_blocks = nn.ModuleList(
|
250 |
+
[BasicTransformerBlock(inner_dim, n_heads, d_head, dropout=dropout, context_dim=context_dim)
|
251 |
+
for d in range(depth)]
|
252 |
+
)
|
253 |
+
|
254 |
+
self.proj_out = zero_module(nn.Conv2d(inner_dim,
|
255 |
+
in_channels,
|
256 |
+
kernel_size=1,
|
257 |
+
stride=1,
|
258 |
+
padding=0))
|
259 |
+
|
260 |
+
def forward(self, x, context=None, attn_dict=None, layer_type=None):
|
261 |
+
# note: if no context is given, cross-attention defaults to self-attention
|
262 |
+
b, c, h, w = x.shape
|
263 |
+
x_in = x
|
264 |
+
x = self.norm(x)
|
265 |
+
x = self.proj_in(x)
|
266 |
+
x = rearrange(x, 'b c h w -> b (h w) c')
|
267 |
+
for block in self.transformer_blocks:
|
268 |
+
x = block(x, context=context, attn_dict=attn_dict, layer_type=layer_type)
|
269 |
+
x = rearrange(x, 'b (h w) c -> b c h w', h=h, w=w)
|
270 |
+
x = self.proj_out(x)
|
271 |
+
return x + x_in
|
ldm/modules/diffusionmodules/model.py
ADDED
@@ -0,0 +1,836 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# pytorch_diffusion + derived encoder decoder
|
2 |
+
import math
|
3 |
+
import torch
|
4 |
+
import torch.nn as nn
|
5 |
+
import numpy as np
|
6 |
+
from einops import rearrange
|
7 |
+
|
8 |
+
from ldm.util import instantiate_from_config
|
9 |
+
from ldm.modules.attention import LinearAttention
|
10 |
+
|
11 |
+
|
12 |
+
def get_timestep_embedding(timesteps, embedding_dim):
|
13 |
+
"""
|
14 |
+
This matches the implementation in Denoising Diffusion Probabilistic Models:
|
15 |
+
From Fairseq.
|
16 |
+
Build sinusoidal embeddings.
|
17 |
+
This matches the implementation in tensor2tensor, but differs slightly
|
18 |
+
from the description in Section 3.5 of "Attention Is All You Need".
|
19 |
+
"""
|
20 |
+
assert len(timesteps.shape) == 1
|
21 |
+
|
22 |
+
half_dim = embedding_dim // 2
|
23 |
+
emb = math.log(10000) / (half_dim - 1)
|
24 |
+
emb = torch.exp(torch.arange(half_dim, dtype=torch.float32) * -emb)
|
25 |
+
emb = emb.to(device=timesteps.device)
|
26 |
+
emb = timesteps.float()[:, None] * emb[None, :]
|
27 |
+
emb = torch.cat([torch.sin(emb), torch.cos(emb)], dim=1)
|
28 |
+
if embedding_dim % 2 == 1: # zero pad
|
29 |
+
emb = torch.nn.functional.pad(emb, (0,1,0,0))
|
30 |
+
return emb
|
31 |
+
|
32 |
+
|
33 |
+
def nonlinearity(x):
|
34 |
+
# swish
|
35 |
+
return x*torch.sigmoid(x)
|
36 |
+
|
37 |
+
|
38 |
+
def Normalize(in_channels, num_groups=32):
|
39 |
+
return torch.nn.GroupNorm(num_groups=num_groups, num_channels=in_channels, eps=1e-6, affine=True)
|
40 |
+
|
41 |
+
|
42 |
+
class Upsample(nn.Module):
|
43 |
+
def __init__(self, in_channels, with_conv):
|
44 |
+
super().__init__()
|
45 |
+
self.with_conv = with_conv
|
46 |
+
if self.with_conv:
|
47 |
+
self.conv = torch.nn.Conv2d(in_channels,
|
48 |
+
in_channels,
|
49 |
+
kernel_size=3,
|
50 |
+
stride=1,
|
51 |
+
padding=1)
|
52 |
+
|
53 |
+
def forward(self, x):
|
54 |
+
x = torch.nn.functional.interpolate(x, scale_factor=2.0, mode="nearest")
|
55 |
+
if self.with_conv:
|
56 |
+
x = self.conv(x)
|
57 |
+
return x
|
58 |
+
|
59 |
+
|
60 |
+
class Downsample(nn.Module):
|
61 |
+
def __init__(self, in_channels, with_conv):
|
62 |
+
super().__init__()
|
63 |
+
self.with_conv = with_conv
|
64 |
+
if self.with_conv:
|
65 |
+
# no asymmetric padding in torch conv, must do it ourselves
|
66 |
+
self.conv = torch.nn.Conv2d(in_channels,
|
67 |
+
in_channels,
|
68 |
+
kernel_size=3,
|
69 |
+
stride=2,
|
70 |
+
padding=0)
|
71 |
+
|
72 |
+
def forward(self, x):
|
73 |
+
if self.with_conv:
|
74 |
+
pad = (0,1,0,1)
|
75 |
+
x = torch.nn.functional.pad(x, pad, mode="constant", value=0)
|
76 |
+
x = self.conv(x)
|
77 |
+
else:
|
78 |
+
x = torch.nn.functional.avg_pool2d(x, kernel_size=2, stride=2)
|
79 |
+
return x
|
80 |
+
|
81 |
+
|
82 |
+
class ResnetBlock(nn.Module):
|
83 |
+
def __init__(self, *, in_channels, out_channels=None, conv_shortcut=False,
|
84 |
+
dropout, temb_channels=512):
|
85 |
+
super().__init__()
|
86 |
+
self.in_channels = in_channels
|
87 |
+
out_channels = in_channels if out_channels is None else out_channels
|
88 |
+
self.out_channels = out_channels
|
89 |
+
self.use_conv_shortcut = conv_shortcut
|
90 |
+
|
91 |
+
self.norm1 = Normalize(in_channels)
|
92 |
+
self.conv1 = torch.nn.Conv2d(in_channels,
|
93 |
+
out_channels,
|
94 |
+
kernel_size=3,
|
95 |
+
stride=1,
|
96 |
+
padding=1)
|
97 |
+
if temb_channels > 0:
|
98 |
+
self.temb_proj = torch.nn.Linear(temb_channels,
|
99 |
+
out_channels)
|
100 |
+
self.norm2 = Normalize(out_channels)
|
101 |
+
self.dropout = torch.nn.Dropout(dropout)
|
102 |
+
self.conv2 = torch.nn.Conv2d(out_channels,
|
103 |
+
out_channels,
|
104 |
+
kernel_size=3,
|
105 |
+
stride=1,
|
106 |
+
padding=1)
|
107 |
+
if self.in_channels != self.out_channels:
|
108 |
+
if self.use_conv_shortcut:
|
109 |
+
self.conv_shortcut = torch.nn.Conv2d(in_channels,
|
110 |
+
out_channels,
|
111 |
+
kernel_size=3,
|
112 |
+
stride=1,
|
113 |
+
padding=1)
|
114 |
+
else:
|
115 |
+
self.nin_shortcut = torch.nn.Conv2d(in_channels,
|
116 |
+
out_channels,
|
117 |
+
kernel_size=1,
|
118 |
+
stride=1,
|
119 |
+
padding=0)
|
120 |
+
|
121 |
+
def forward(self, x, temb):
|
122 |
+
h = x
|
123 |
+
h = self.norm1(h)
|
124 |
+
h = nonlinearity(h)
|
125 |
+
h = self.conv1(h)
|
126 |
+
|
127 |
+
if temb is not None:
|
128 |
+
h = h + self.temb_proj(nonlinearity(temb))[:,:,None,None]
|
129 |
+
|
130 |
+
h = self.norm2(h)
|
131 |
+
h = nonlinearity(h)
|
132 |
+
h = self.dropout(h)
|
133 |
+
h = self.conv2(h)
|
134 |
+
|
135 |
+
if self.in_channels != self.out_channels:
|
136 |
+
if self.use_conv_shortcut:
|
137 |
+
x = self.conv_shortcut(x)
|
138 |
+
else:
|
139 |
+
x = self.nin_shortcut(x)
|
140 |
+
|
141 |
+
return x+h
|
142 |
+
|
143 |
+
|
144 |
+
class LinAttnBlock(LinearAttention):
|
145 |
+
"""to match AttnBlock usage"""
|
146 |
+
def __init__(self, in_channels):
|
147 |
+
super().__init__(dim=in_channels, heads=1, dim_head=in_channels)
|
148 |
+
|
149 |
+
|
150 |
+
class AttnBlock(nn.Module):
|
151 |
+
def __init__(self, in_channels):
|
152 |
+
super().__init__()
|
153 |
+
self.in_channels = in_channels
|
154 |
+
|
155 |
+
self.norm = Normalize(in_channels)
|
156 |
+
self.q = torch.nn.Conv2d(in_channels,
|
157 |
+
in_channels,
|
158 |
+
kernel_size=1,
|
159 |
+
stride=1,
|
160 |
+
padding=0)
|
161 |
+
self.k = torch.nn.Conv2d(in_channels,
|
162 |
+
in_channels,
|
163 |
+
kernel_size=1,
|
164 |
+
stride=1,
|
165 |
+
padding=0)
|
166 |
+
self.v = torch.nn.Conv2d(in_channels,
|
167 |
+
in_channels,
|
168 |
+
kernel_size=1,
|
169 |
+
stride=1,
|
170 |
+
padding=0)
|
171 |
+
self.proj_out = torch.nn.Conv2d(in_channels,
|
172 |
+
in_channels,
|
173 |
+
kernel_size=1,
|
174 |
+
stride=1,
|
175 |
+
padding=0)
|
176 |
+
|
177 |
+
|
178 |
+
def forward(self, x):
|
179 |
+
h_ = x
|
180 |
+
h_ = self.norm(h_)
|
181 |
+
q = self.q(h_)
|
182 |
+
k = self.k(h_)
|
183 |
+
v = self.v(h_)
|
184 |
+
|
185 |
+
# compute attention
|
186 |
+
b,c,h,w = q.shape
|
187 |
+
q = q.reshape(b,c,h*w)
|
188 |
+
q = q.permute(0,2,1) # b,hw,c
|
189 |
+
k = k.reshape(b,c,h*w) # b,c,hw
|
190 |
+
w_ = torch.bmm(q,k) # b,hw,hw w[b,i,j]=sum_c q[b,i,c]k[b,c,j]
|
191 |
+
w_ = w_ * (int(c)**(-0.5))
|
192 |
+
w_ = torch.nn.functional.softmax(w_, dim=2)
|
193 |
+
|
194 |
+
# attend to values
|
195 |
+
v = v.reshape(b,c,h*w)
|
196 |
+
w_ = w_.permute(0,2,1) # b,hw,hw (first hw of k, second of q)
|
197 |
+
h_ = torch.bmm(v,w_) # b, c,hw (hw of q) h_[b,c,j] = sum_i v[b,c,i] w_[b,i,j]
|
198 |
+
h_ = h_.reshape(b,c,h,w)
|
199 |
+
|
200 |
+
h_ = self.proj_out(h_)
|
201 |
+
|
202 |
+
return x+h_
|
203 |
+
|
204 |
+
|
205 |
+
def make_attn(in_channels, attn_type="vanilla"):
|
206 |
+
assert attn_type in ["vanilla", "linear", "none"], f'attn_type {attn_type} unknown'
|
207 |
+
print(f"making attention of type '{attn_type}' with {in_channels} in_channels")
|
208 |
+
if attn_type == "vanilla":
|
209 |
+
return AttnBlock(in_channels)
|
210 |
+
elif attn_type == "none":
|
211 |
+
return nn.Identity(in_channels)
|
212 |
+
else:
|
213 |
+
return LinAttnBlock(in_channels)
|
214 |
+
|
215 |
+
|
216 |
+
class Model(nn.Module):
|
217 |
+
def __init__(self, *, ch, out_ch, ch_mult=(1,2,4,8), num_res_blocks,
|
218 |
+
attn_resolutions, dropout=0.0, resamp_with_conv=True, in_channels,
|
219 |
+
resolution, use_timestep=True, use_linear_attn=False, attn_type="vanilla"):
|
220 |
+
super().__init__()
|
221 |
+
if use_linear_attn: attn_type = "linear"
|
222 |
+
self.ch = ch
|
223 |
+
self.temb_ch = self.ch*4
|
224 |
+
self.num_resolutions = len(ch_mult)
|
225 |
+
self.num_res_blocks = num_res_blocks
|
226 |
+
self.resolution = resolution
|
227 |
+
self.in_channels = in_channels
|
228 |
+
|
229 |
+
self.use_timestep = use_timestep
|
230 |
+
if self.use_timestep:
|
231 |
+
# timestep embedding
|
232 |
+
self.temb = nn.Module()
|
233 |
+
self.temb.dense = nn.ModuleList([
|
234 |
+
torch.nn.Linear(self.ch,
|
235 |
+
self.temb_ch),
|
236 |
+
torch.nn.Linear(self.temb_ch,
|
237 |
+
self.temb_ch),
|
238 |
+
])
|
239 |
+
|
240 |
+
# downsampling
|
241 |
+
self.conv_in = torch.nn.Conv2d(in_channels,
|
242 |
+
self.ch,
|
243 |
+
kernel_size=3,
|
244 |
+
stride=1,
|
245 |
+
padding=1)
|
246 |
+
|
247 |
+
curr_res = resolution
|
248 |
+
in_ch_mult = (1,)+tuple(ch_mult)
|
249 |
+
self.down = nn.ModuleList()
|
250 |
+
for i_level in range(self.num_resolutions):
|
251 |
+
block = nn.ModuleList()
|
252 |
+
attn = nn.ModuleList()
|
253 |
+
block_in = ch*in_ch_mult[i_level]
|
254 |
+
block_out = ch*ch_mult[i_level]
|
255 |
+
for i_block in range(self.num_res_blocks):
|
256 |
+
block.append(ResnetBlock(in_channels=block_in,
|
257 |
+
out_channels=block_out,
|
258 |
+
temb_channels=self.temb_ch,
|
259 |
+
dropout=dropout))
|
260 |
+
block_in = block_out
|
261 |
+
if curr_res in attn_resolutions:
|
262 |
+
attn.append(make_attn(block_in, attn_type=attn_type))
|
263 |
+
down = nn.Module()
|
264 |
+
down.block = block
|
265 |
+
down.attn = attn
|
266 |
+
if i_level != self.num_resolutions-1:
|
267 |
+
down.downsample = Downsample(block_in, resamp_with_conv)
|
268 |
+
curr_res = curr_res // 2
|
269 |
+
self.down.append(down)
|
270 |
+
|
271 |
+
# middle
|
272 |
+
self.mid = nn.Module()
|
273 |
+
self.mid.block_1 = ResnetBlock(in_channels=block_in,
|
274 |
+
out_channels=block_in,
|
275 |
+
temb_channels=self.temb_ch,
|
276 |
+
dropout=dropout)
|
277 |
+
self.mid.attn_1 = make_attn(block_in, attn_type=attn_type)
|
278 |
+
self.mid.block_2 = ResnetBlock(in_channels=block_in,
|
279 |
+
out_channels=block_in,
|
280 |
+
temb_channels=self.temb_ch,
|
281 |
+
dropout=dropout)
|
282 |
+
|
283 |
+
# upsampling
|
284 |
+
self.up = nn.ModuleList()
|
285 |
+
for i_level in reversed(range(self.num_resolutions)):
|
286 |
+
block = nn.ModuleList()
|
287 |
+
attn = nn.ModuleList()
|
288 |
+
block_out = ch*ch_mult[i_level]
|
289 |
+
skip_in = ch*ch_mult[i_level]
|
290 |
+
for i_block in range(self.num_res_blocks+1):
|
291 |
+
if i_block == self.num_res_blocks:
|
292 |
+
skip_in = ch*in_ch_mult[i_level]
|
293 |
+
block.append(ResnetBlock(in_channels=block_in+skip_in,
|
294 |
+
out_channels=block_out,
|
295 |
+
temb_channels=self.temb_ch,
|
296 |
+
dropout=dropout))
|
297 |
+
block_in = block_out
|
298 |
+
if curr_res in attn_resolutions:
|
299 |
+
attn.append(make_attn(block_in, attn_type=attn_type))
|
300 |
+
up = nn.Module()
|
301 |
+
up.block = block
|
302 |
+
up.attn = attn
|
303 |
+
if i_level != 0:
|
304 |
+
up.upsample = Upsample(block_in, resamp_with_conv)
|
305 |
+
curr_res = curr_res * 2
|
306 |
+
self.up.insert(0, up) # prepend to get consistent order
|
307 |
+
|
308 |
+
# end
|
309 |
+
self.norm_out = Normalize(block_in)
|
310 |
+
self.conv_out = torch.nn.Conv2d(block_in,
|
311 |
+
out_ch,
|
312 |
+
kernel_size=3,
|
313 |
+
stride=1,
|
314 |
+
padding=1)
|
315 |
+
|
316 |
+
def forward(self, x, t=None, context=None):
|
317 |
+
#assert x.shape[2] == x.shape[3] == self.resolution
|
318 |
+
if context is not None:
|
319 |
+
# assume aligned context, cat along channel axis
|
320 |
+
x = torch.cat((x, context), dim=1)
|
321 |
+
if self.use_timestep:
|
322 |
+
# timestep embedding
|
323 |
+
assert t is not None
|
324 |
+
temb = get_timestep_embedding(t, self.ch)
|
325 |
+
temb = self.temb.dense[0](temb)
|
326 |
+
temb = nonlinearity(temb)
|
327 |
+
temb = self.temb.dense[1](temb)
|
328 |
+
else:
|
329 |
+
temb = None
|
330 |
+
|
331 |
+
# downsampling
|
332 |
+
hs = [self.conv_in(x)]
|
333 |
+
for i_level in range(self.num_resolutions):
|
334 |
+
for i_block in range(self.num_res_blocks):
|
335 |
+
h = self.down[i_level].block[i_block](hs[-1], temb)
|
336 |
+
if len(self.down[i_level].attn) > 0:
|
337 |
+
h = self.down[i_level].attn[i_block](h)
|
338 |
+
hs.append(h)
|
339 |
+
if i_level != self.num_resolutions-1:
|
340 |
+
hs.append(self.down[i_level].downsample(hs[-1]))
|
341 |
+
|
342 |
+
# middle
|
343 |
+
h = hs[-1]
|
344 |
+
h = self.mid.block_1(h, temb)
|
345 |
+
h = self.mid.attn_1(h)
|
346 |
+
h = self.mid.block_2(h, temb)
|
347 |
+
|
348 |
+
# upsampling
|
349 |
+
for i_level in reversed(range(self.num_resolutions)):
|
350 |
+
for i_block in range(self.num_res_blocks+1):
|
351 |
+
h = self.up[i_level].block[i_block](
|
352 |
+
torch.cat([h, hs.pop()], dim=1), temb)
|
353 |
+
if len(self.up[i_level].attn) > 0:
|
354 |
+
h = self.up[i_level].attn[i_block](h)
|
355 |
+
if i_level != 0:
|
356 |
+
h = self.up[i_level].upsample(h)
|
357 |
+
|
358 |
+
# end
|
359 |
+
h = self.norm_out(h)
|
360 |
+
h = nonlinearity(h)
|
361 |
+
h = self.conv_out(h)
|
362 |
+
return h
|
363 |
+
|
364 |
+
def get_last_layer(self):
|
365 |
+
return self.conv_out.weight
|
366 |
+
|
367 |
+
|
368 |
+
class Encoder(nn.Module):
|
369 |
+
def __init__(self, *, ch, out_ch, ch_mult=(1,2,4,8), num_res_blocks,
|
370 |
+
attn_resolutions, dropout=0.0, resamp_with_conv=True, in_channels,
|
371 |
+
resolution, z_channels, double_z=True, use_linear_attn=False, attn_type="vanilla",
|
372 |
+
**ignore_kwargs):
|
373 |
+
super().__init__()
|
374 |
+
if use_linear_attn: attn_type = "linear"
|
375 |
+
self.ch = ch
|
376 |
+
self.temb_ch = 0
|
377 |
+
self.num_resolutions = len(ch_mult)
|
378 |
+
self.num_res_blocks = num_res_blocks
|
379 |
+
self.resolution = resolution
|
380 |
+
self.in_channels = in_channels
|
381 |
+
|
382 |
+
# downsampling
|
383 |
+
self.conv_in = torch.nn.Conv2d(in_channels,
|
384 |
+
self.ch,
|
385 |
+
kernel_size=3,
|
386 |
+
stride=1,
|
387 |
+
padding=1)
|
388 |
+
|
389 |
+
curr_res = resolution
|
390 |
+
in_ch_mult = (1,)+tuple(ch_mult)
|
391 |
+
self.in_ch_mult = in_ch_mult
|
392 |
+
self.down = nn.ModuleList()
|
393 |
+
for i_level in range(self.num_resolutions):
|
394 |
+
block = nn.ModuleList()
|
395 |
+
attn = nn.ModuleList()
|
396 |
+
block_in = ch*in_ch_mult[i_level]
|
397 |
+
block_out = ch*ch_mult[i_level]
|
398 |
+
for i_block in range(self.num_res_blocks):
|
399 |
+
block.append(ResnetBlock(in_channels=block_in,
|
400 |
+
out_channels=block_out,
|
401 |
+
temb_channels=self.temb_ch,
|
402 |
+
dropout=dropout))
|
403 |
+
block_in = block_out
|
404 |
+
if curr_res in attn_resolutions:
|
405 |
+
attn.append(make_attn(block_in, attn_type=attn_type))
|
406 |
+
down = nn.Module()
|
407 |
+
down.block = block
|
408 |
+
down.attn = attn
|
409 |
+
if i_level != self.num_resolutions-1:
|
410 |
+
down.downsample = Downsample(block_in, resamp_with_conv)
|
411 |
+
curr_res = curr_res // 2
|
412 |
+
self.down.append(down)
|
413 |
+
|
414 |
+
# middle
|
415 |
+
self.mid = nn.Module()
|
416 |
+
self.mid.block_1 = ResnetBlock(in_channels=block_in,
|
417 |
+
out_channels=block_in,
|
418 |
+
temb_channels=self.temb_ch,
|
419 |
+
dropout=dropout)
|
420 |
+
self.mid.attn_1 = make_attn(block_in, attn_type=attn_type)
|
421 |
+
self.mid.block_2 = ResnetBlock(in_channels=block_in,
|
422 |
+
out_channels=block_in,
|
423 |
+
temb_channels=self.temb_ch,
|
424 |
+
dropout=dropout)
|
425 |
+
|
426 |
+
# end
|
427 |
+
self.norm_out = Normalize(block_in)
|
428 |
+
self.conv_out = torch.nn.Conv2d(block_in,
|
429 |
+
2*z_channels if double_z else z_channels,
|
430 |
+
kernel_size=3,
|
431 |
+
stride=1,
|
432 |
+
padding=1)
|
433 |
+
|
434 |
+
def forward(self, x):
|
435 |
+
# timestep embedding
|
436 |
+
temb = None
|
437 |
+
|
438 |
+
# downsampling
|
439 |
+
hs = [self.conv_in(x)]
|
440 |
+
for i_level in range(self.num_resolutions):
|
441 |
+
for i_block in range(self.num_res_blocks):
|
442 |
+
h = self.down[i_level].block[i_block](hs[-1], temb)
|
443 |
+
if len(self.down[i_level].attn) > 0:
|
444 |
+
h = self.down[i_level].attn[i_block](h)
|
445 |
+
hs.append(h)
|
446 |
+
if i_level != self.num_resolutions-1:
|
447 |
+
hs.append(self.down[i_level].downsample(hs[-1]))
|
448 |
+
|
449 |
+
# middle
|
450 |
+
h = hs[-1]
|
451 |
+
h = self.mid.block_1(h, temb)
|
452 |
+
h = self.mid.attn_1(h)
|
453 |
+
h = self.mid.block_2(h, temb)
|
454 |
+
|
455 |
+
# end
|
456 |
+
h = self.norm_out(h)
|
457 |
+
h = nonlinearity(h)
|
458 |
+
h = self.conv_out(h)
|
459 |
+
return h
|
460 |
+
|
461 |
+
|
462 |
+
class Decoder(nn.Module):
|
463 |
+
def __init__(self, *, ch, out_ch, ch_mult=(1,2,4,8), num_res_blocks,
|
464 |
+
attn_resolutions, dropout=0.0, resamp_with_conv=True, in_channels,
|
465 |
+
resolution, z_channels, give_pre_end=False, tanh_out=False, use_linear_attn=False,
|
466 |
+
attn_type="vanilla", **ignorekwargs):
|
467 |
+
super().__init__()
|
468 |
+
if use_linear_attn: attn_type = "linear"
|
469 |
+
self.ch = ch
|
470 |
+
self.temb_ch = 0
|
471 |
+
self.num_resolutions = len(ch_mult)
|
472 |
+
self.num_res_blocks = num_res_blocks
|
473 |
+
self.resolution = resolution
|
474 |
+
self.in_channels = in_channels
|
475 |
+
self.give_pre_end = give_pre_end
|
476 |
+
self.tanh_out = tanh_out
|
477 |
+
|
478 |
+
# compute in_ch_mult, block_in and curr_res at lowest res
|
479 |
+
in_ch_mult = (1,)+tuple(ch_mult)
|
480 |
+
block_in = ch*ch_mult[self.num_resolutions-1]
|
481 |
+
curr_res = resolution // 2**(self.num_resolutions-1)
|
482 |
+
self.z_shape = (1,z_channels,curr_res,curr_res)
|
483 |
+
print("Working with z of shape {} = {} dimensions.".format(
|
484 |
+
self.z_shape, np.prod(self.z_shape)))
|
485 |
+
|
486 |
+
# z to block_in
|
487 |
+
self.conv_in = torch.nn.Conv2d(z_channels,
|
488 |
+
block_in,
|
489 |
+
kernel_size=3,
|
490 |
+
stride=1,
|
491 |
+
padding=1)
|
492 |
+
|
493 |
+
# middle
|
494 |
+
self.mid = nn.Module()
|
495 |
+
self.mid.block_1 = ResnetBlock(in_channels=block_in,
|
496 |
+
out_channels=block_in,
|
497 |
+
temb_channels=self.temb_ch,
|
498 |
+
dropout=dropout)
|
499 |
+
self.mid.attn_1 = make_attn(block_in, attn_type=attn_type)
|
500 |
+
self.mid.block_2 = ResnetBlock(in_channels=block_in,
|
501 |
+
out_channels=block_in,
|
502 |
+
temb_channels=self.temb_ch,
|
503 |
+
dropout=dropout)
|
504 |
+
|
505 |
+
# upsampling
|
506 |
+
self.up = nn.ModuleList()
|
507 |
+
for i_level in reversed(range(self.num_resolutions)):
|
508 |
+
block = nn.ModuleList()
|
509 |
+
attn = nn.ModuleList()
|
510 |
+
block_out = ch*ch_mult[i_level]
|
511 |
+
for i_block in range(self.num_res_blocks+1):
|
512 |
+
block.append(ResnetBlock(in_channels=block_in,
|
513 |
+
out_channels=block_out,
|
514 |
+
temb_channels=self.temb_ch,
|
515 |
+
dropout=dropout))
|
516 |
+
block_in = block_out
|
517 |
+
if curr_res in attn_resolutions:
|
518 |
+
attn.append(make_attn(block_in, attn_type=attn_type))
|
519 |
+
up = nn.Module()
|
520 |
+
up.block = block
|
521 |
+
up.attn = attn
|
522 |
+
if i_level != 0:
|
523 |
+
up.upsample = Upsample(block_in, resamp_with_conv)
|
524 |
+
curr_res = curr_res * 2
|
525 |
+
self.up.insert(0, up) # prepend to get consistent order
|
526 |
+
|
527 |
+
# end
|
528 |
+
self.norm_out = Normalize(block_in)
|
529 |
+
self.conv_out = torch.nn.Conv2d(block_in,
|
530 |
+
out_ch,
|
531 |
+
kernel_size=3,
|
532 |
+
stride=1,
|
533 |
+
padding=1)
|
534 |
+
|
535 |
+
def forward(self, z):
|
536 |
+
|
537 |
+
#assert z.shape[1:] == self.z_shape[1:]
|
538 |
+
self.last_z_shape = z.shape
|
539 |
+
|
540 |
+
# timestep embedding
|
541 |
+
temb = None
|
542 |
+
|
543 |
+
# z to block_in
|
544 |
+
h = self.conv_in(z)
|
545 |
+
|
546 |
+
# middle
|
547 |
+
h = self.mid.block_1(h, temb)
|
548 |
+
h = self.mid.attn_1(h)
|
549 |
+
h = self.mid.block_2(h, temb)
|
550 |
+
|
551 |
+
# upsampling
|
552 |
+
for i_level in reversed(range(self.num_resolutions)):
|
553 |
+
for i_block in range(self.num_res_blocks+1):
|
554 |
+
h = self.up[i_level].block[i_block](h, temb)
|
555 |
+
if len(self.up[i_level].attn) > 0:
|
556 |
+
h = self.up[i_level].attn[i_block](h)
|
557 |
+
if i_level != 0:
|
558 |
+
h = self.up[i_level].upsample(h)
|
559 |
+
|
560 |
+
# end
|
561 |
+
if self.give_pre_end:
|
562 |
+
return h
|
563 |
+
|
564 |
+
h = self.norm_out(h)
|
565 |
+
h = nonlinearity(h)
|
566 |
+
h = self.conv_out(h)
|
567 |
+
if self.tanh_out:
|
568 |
+
h = torch.tanh(h)
|
569 |
+
return h
|
570 |
+
|
571 |
+
|
572 |
+
class SimpleDecoder(nn.Module):
|
573 |
+
def __init__(self, in_channels, out_channels, *args, **kwargs):
|
574 |
+
super().__init__()
|
575 |
+
self.model = nn.ModuleList([nn.Conv2d(in_channels, in_channels, 1),
|
576 |
+
ResnetBlock(in_channels=in_channels,
|
577 |
+
out_channels=2 * in_channels,
|
578 |
+
temb_channels=0, dropout=0.0),
|
579 |
+
ResnetBlock(in_channels=2 * in_channels,
|
580 |
+
out_channels=4 * in_channels,
|
581 |
+
temb_channels=0, dropout=0.0),
|
582 |
+
ResnetBlock(in_channels=4 * in_channels,
|
583 |
+
out_channels=2 * in_channels,
|
584 |
+
temb_channels=0, dropout=0.0),
|
585 |
+
nn.Conv2d(2*in_channels, in_channels, 1),
|
586 |
+
Upsample(in_channels, with_conv=True)])
|
587 |
+
# end
|
588 |
+
self.norm_out = Normalize(in_channels)
|
589 |
+
self.conv_out = torch.nn.Conv2d(in_channels,
|
590 |
+
out_channels,
|
591 |
+
kernel_size=3,
|
592 |
+
stride=1,
|
593 |
+
padding=1)
|
594 |
+
|
595 |
+
def forward(self, x):
|
596 |
+
for i, layer in enumerate(self.model):
|
597 |
+
if i in [1,2,3]:
|
598 |
+
x = layer(x, None)
|
599 |
+
else:
|
600 |
+
x = layer(x)
|
601 |
+
|
602 |
+
h = self.norm_out(x)
|
603 |
+
h = nonlinearity(h)
|
604 |
+
x = self.conv_out(h)
|
605 |
+
return x
|
606 |
+
|
607 |
+
|
608 |
+
class UpsampleDecoder(nn.Module):
|
609 |
+
def __init__(self, in_channels, out_channels, ch, num_res_blocks, resolution,
|
610 |
+
ch_mult=(2,2), dropout=0.0):
|
611 |
+
super().__init__()
|
612 |
+
# upsampling
|
613 |
+
self.temb_ch = 0
|
614 |
+
self.num_resolutions = len(ch_mult)
|
615 |
+
self.num_res_blocks = num_res_blocks
|
616 |
+
block_in = in_channels
|
617 |
+
curr_res = resolution // 2 ** (self.num_resolutions - 1)
|
618 |
+
self.res_blocks = nn.ModuleList()
|
619 |
+
self.upsample_blocks = nn.ModuleList()
|
620 |
+
for i_level in range(self.num_resolutions):
|
621 |
+
res_block = []
|
622 |
+
block_out = ch * ch_mult[i_level]
|
623 |
+
for i_block in range(self.num_res_blocks + 1):
|
624 |
+
res_block.append(ResnetBlock(in_channels=block_in,
|
625 |
+
out_channels=block_out,
|
626 |
+
temb_channels=self.temb_ch,
|
627 |
+
dropout=dropout))
|
628 |
+
block_in = block_out
|
629 |
+
self.res_blocks.append(nn.ModuleList(res_block))
|
630 |
+
if i_level != self.num_resolutions - 1:
|
631 |
+
self.upsample_blocks.append(Upsample(block_in, True))
|
632 |
+
curr_res = curr_res * 2
|
633 |
+
|
634 |
+
# end
|
635 |
+
self.norm_out = Normalize(block_in)
|
636 |
+
self.conv_out = torch.nn.Conv2d(block_in,
|
637 |
+
out_channels,
|
638 |
+
kernel_size=3,
|
639 |
+
stride=1,
|
640 |
+
padding=1)
|
641 |
+
|
642 |
+
def forward(self, x):
|
643 |
+
# upsampling
|
644 |
+
h = x
|
645 |
+
for k, i_level in enumerate(range(self.num_resolutions)):
|
646 |
+
for i_block in range(self.num_res_blocks + 1):
|
647 |
+
h = self.res_blocks[i_level][i_block](h, None)
|
648 |
+
if i_level != self.num_resolutions - 1:
|
649 |
+
h = self.upsample_blocks[k](h)
|
650 |
+
h = self.norm_out(h)
|
651 |
+
h = nonlinearity(h)
|
652 |
+
h = self.conv_out(h)
|
653 |
+
return h
|
654 |
+
|
655 |
+
|
656 |
+
class LatentRescaler(nn.Module):
|
657 |
+
def __init__(self, factor, in_channels, mid_channels, out_channels, depth=2):
|
658 |
+
super().__init__()
|
659 |
+
# residual block, interpolate, residual block
|
660 |
+
self.factor = factor
|
661 |
+
self.conv_in = nn.Conv2d(in_channels,
|
662 |
+
mid_channels,
|
663 |
+
kernel_size=3,
|
664 |
+
stride=1,
|
665 |
+
padding=1)
|
666 |
+
self.res_block1 = nn.ModuleList([ResnetBlock(in_channels=mid_channels,
|
667 |
+
out_channels=mid_channels,
|
668 |
+
temb_channels=0,
|
669 |
+
dropout=0.0) for _ in range(depth)])
|
670 |
+
self.attn = AttnBlock(mid_channels)
|
671 |
+
self.res_block2 = nn.ModuleList([ResnetBlock(in_channels=mid_channels,
|
672 |
+
out_channels=mid_channels,
|
673 |
+
temb_channels=0,
|
674 |
+
dropout=0.0) for _ in range(depth)])
|
675 |
+
|
676 |
+
self.conv_out = nn.Conv2d(mid_channels,
|
677 |
+
out_channels,
|
678 |
+
kernel_size=1,
|
679 |
+
)
|
680 |
+
|
681 |
+
def forward(self, x):
|
682 |
+
x = self.conv_in(x)
|
683 |
+
for block in self.res_block1:
|
684 |
+
x = block(x, None)
|
685 |
+
x = torch.nn.functional.interpolate(x, size=(int(round(x.shape[2]*self.factor)), int(round(x.shape[3]*self.factor))))
|
686 |
+
x = self.attn(x)
|
687 |
+
for block in self.res_block2:
|
688 |
+
x = block(x, None)
|
689 |
+
x = self.conv_out(x)
|
690 |
+
return x
|
691 |
+
|
692 |
+
|
693 |
+
class MergedRescaleEncoder(nn.Module):
|
694 |
+
def __init__(self, in_channels, ch, resolution, out_ch, num_res_blocks,
|
695 |
+
attn_resolutions, dropout=0.0, resamp_with_conv=True,
|
696 |
+
ch_mult=(1,2,4,8), rescale_factor=1.0, rescale_module_depth=1):
|
697 |
+
super().__init__()
|
698 |
+
intermediate_chn = ch * ch_mult[-1]
|
699 |
+
self.encoder = Encoder(in_channels=in_channels, num_res_blocks=num_res_blocks, ch=ch, ch_mult=ch_mult,
|
700 |
+
z_channels=intermediate_chn, double_z=False, resolution=resolution,
|
701 |
+
attn_resolutions=attn_resolutions, dropout=dropout, resamp_with_conv=resamp_with_conv,
|
702 |
+
out_ch=None)
|
703 |
+
self.rescaler = LatentRescaler(factor=rescale_factor, in_channels=intermediate_chn,
|
704 |
+
mid_channels=intermediate_chn, out_channels=out_ch, depth=rescale_module_depth)
|
705 |
+
|
706 |
+
def forward(self, x):
|
707 |
+
x = self.encoder(x)
|
708 |
+
x = self.rescaler(x)
|
709 |
+
return x
|
710 |
+
|
711 |
+
|
712 |
+
class MergedRescaleDecoder(nn.Module):
|
713 |
+
def __init__(self, z_channels, out_ch, resolution, num_res_blocks, attn_resolutions, ch, ch_mult=(1,2,4,8),
|
714 |
+
dropout=0.0, resamp_with_conv=True, rescale_factor=1.0, rescale_module_depth=1):
|
715 |
+
super().__init__()
|
716 |
+
tmp_chn = z_channels*ch_mult[-1]
|
717 |
+
self.decoder = Decoder(out_ch=out_ch, z_channels=tmp_chn, attn_resolutions=attn_resolutions, dropout=dropout,
|
718 |
+
resamp_with_conv=resamp_with_conv, in_channels=None, num_res_blocks=num_res_blocks,
|
719 |
+
ch_mult=ch_mult, resolution=resolution, ch=ch)
|
720 |
+
self.rescaler = LatentRescaler(factor=rescale_factor, in_channels=z_channels, mid_channels=tmp_chn,
|
721 |
+
out_channels=tmp_chn, depth=rescale_module_depth)
|
722 |
+
|
723 |
+
def forward(self, x):
|
724 |
+
x = self.rescaler(x)
|
725 |
+
x = self.decoder(x)
|
726 |
+
return x
|
727 |
+
|
728 |
+
|
729 |
+
class Upsampler(nn.Module):
|
730 |
+
def __init__(self, in_size, out_size, in_channels, out_channels, ch_mult=2):
|
731 |
+
super().__init__()
|
732 |
+
assert out_size >= in_size
|
733 |
+
num_blocks = int(np.log2(out_size//in_size))+1
|
734 |
+
factor_up = 1.+ (out_size % in_size)
|
735 |
+
print(f"Building {self.__class__.__name__} with in_size: {in_size} --> out_size {out_size} and factor {factor_up}")
|
736 |
+
self.rescaler = LatentRescaler(factor=factor_up, in_channels=in_channels, mid_channels=2*in_channels,
|
737 |
+
out_channels=in_channels)
|
738 |
+
self.decoder = Decoder(out_ch=out_channels, resolution=out_size, z_channels=in_channels, num_res_blocks=2,
|
739 |
+
attn_resolutions=[], in_channels=None, ch=in_channels,
|
740 |
+
ch_mult=[ch_mult for _ in range(num_blocks)])
|
741 |
+
|
742 |
+
def forward(self, x):
|
743 |
+
x = self.rescaler(x)
|
744 |
+
x = self.decoder(x)
|
745 |
+
return x
|
746 |
+
|
747 |
+
|
748 |
+
class Resize(nn.Module):
|
749 |
+
def __init__(self, in_channels=None, learned=False, mode="bilinear"):
|
750 |
+
super().__init__()
|
751 |
+
self.with_conv = learned
|
752 |
+
self.mode = mode
|
753 |
+
if self.with_conv:
|
754 |
+
print(f"Note: {self.__class__.__name} uses learned downsampling and will ignore the fixed {mode} mode")
|
755 |
+
raise NotImplementedError()
|
756 |
+
assert in_channels is not None
|
757 |
+
# no asymmetric padding in torch conv, must do it ourselves
|
758 |
+
self.conv = torch.nn.Conv2d(in_channels,
|
759 |
+
in_channels,
|
760 |
+
kernel_size=4,
|
761 |
+
stride=2,
|
762 |
+
padding=1)
|
763 |
+
|
764 |
+
def forward(self, x, scale_factor=1.0):
|
765 |
+
if scale_factor==1.0:
|
766 |
+
return x
|
767 |
+
else:
|
768 |
+
x = torch.nn.functional.interpolate(x, mode=self.mode, align_corners=False, scale_factor=scale_factor)
|
769 |
+
return x
|
770 |
+
|
771 |
+
class FirstStagePostProcessor(nn.Module):
|
772 |
+
|
773 |
+
def __init__(self, ch_mult:list, in_channels,
|
774 |
+
pretrained_model:nn.Module=None,
|
775 |
+
reshape=False,
|
776 |
+
n_channels=None,
|
777 |
+
dropout=0.,
|
778 |
+
pretrained_config=None):
|
779 |
+
super().__init__()
|
780 |
+
if pretrained_config is None:
|
781 |
+
assert pretrained_model is not None, 'Either "pretrained_model" or "pretrained_config" must not be None'
|
782 |
+
self.pretrained_model = pretrained_model
|
783 |
+
else:
|
784 |
+
assert pretrained_config is not None, 'Either "pretrained_model" or "pretrained_config" must not be None'
|
785 |
+
self.instantiate_pretrained(pretrained_config)
|
786 |
+
|
787 |
+
self.do_reshape = reshape
|
788 |
+
|
789 |
+
if n_channels is None:
|
790 |
+
n_channels = self.pretrained_model.encoder.ch
|
791 |
+
|
792 |
+
self.proj_norm = Normalize(in_channels,num_groups=in_channels//2)
|
793 |
+
self.proj = nn.Conv2d(in_channels,n_channels,kernel_size=3,
|
794 |
+
stride=1,padding=1)
|
795 |
+
|
796 |
+
blocks = []
|
797 |
+
downs = []
|
798 |
+
ch_in = n_channels
|
799 |
+
for m in ch_mult:
|
800 |
+
blocks.append(ResnetBlock(in_channels=ch_in,out_channels=m*n_channels,dropout=dropout))
|
801 |
+
ch_in = m * n_channels
|
802 |
+
downs.append(Downsample(ch_in, with_conv=False))
|
803 |
+
|
804 |
+
self.model = nn.ModuleList(blocks)
|
805 |
+
self.downsampler = nn.ModuleList(downs)
|
806 |
+
|
807 |
+
|
808 |
+
def instantiate_pretrained(self, config):
|
809 |
+
model = instantiate_from_config(config)
|
810 |
+
self.pretrained_model = model.eval()
|
811 |
+
# self.pretrained_model.train = False
|
812 |
+
for param in self.pretrained_model.parameters():
|
813 |
+
param.requires_grad = False
|
814 |
+
|
815 |
+
|
816 |
+
@torch.no_grad()
|
817 |
+
def encode_with_pretrained(self,x):
|
818 |
+
c = self.pretrained_model.encode(x)
|
819 |
+
if isinstance(c, DiagonalGaussianDistribution):
|
820 |
+
c = c.mode()
|
821 |
+
return c
|
822 |
+
|
823 |
+
def forward(self,x):
|
824 |
+
z_fs = self.encode_with_pretrained(x)
|
825 |
+
z = self.proj_norm(z_fs)
|
826 |
+
z = self.proj(z)
|
827 |
+
z = nonlinearity(z)
|
828 |
+
|
829 |
+
for submodel, downmodel in zip(self.model,self.downsampler):
|
830 |
+
z = submodel(z,temb=None)
|
831 |
+
z = downmodel(z)
|
832 |
+
|
833 |
+
if self.do_reshape:
|
834 |
+
z = rearrange(z,'b c h w -> b (h w) c')
|
835 |
+
return z
|
836 |
+
|
ldm/modules/diffusionmodules/openaimodel.py
ADDED
@@ -0,0 +1,973 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from abc import abstractmethod
|
2 |
+
from functools import partial
|
3 |
+
import math
|
4 |
+
from typing import Iterable
|
5 |
+
|
6 |
+
import numpy as np
|
7 |
+
import torch as th
|
8 |
+
import torch.nn as nn
|
9 |
+
import torch.nn.functional as F
|
10 |
+
|
11 |
+
from ldm.modules.diffusionmodules.util import (
|
12 |
+
checkpoint,
|
13 |
+
conv_nd,
|
14 |
+
linear,
|
15 |
+
avg_pool_nd,
|
16 |
+
zero_module,
|
17 |
+
normalization,
|
18 |
+
timestep_embedding,
|
19 |
+
)
|
20 |
+
from ldm.modules.attention import SpatialTransformer
|
21 |
+
|
22 |
+
|
23 |
+
# dummy replace
|
24 |
+
def convert_module_to_f16(x):
|
25 |
+
pass
|
26 |
+
|
27 |
+
def convert_module_to_f32(x):
|
28 |
+
pass
|
29 |
+
|
30 |
+
|
31 |
+
## go
|
32 |
+
class AttentionPool2d(nn.Module):
|
33 |
+
"""
|
34 |
+
Adapted from CLIP: https://github.com/openai/CLIP/blob/main/clip/model.py
|
35 |
+
"""
|
36 |
+
|
37 |
+
def __init__(
|
38 |
+
self,
|
39 |
+
spacial_dim: int,
|
40 |
+
embed_dim: int,
|
41 |
+
num_heads_channels: int,
|
42 |
+
output_dim: int = None,
|
43 |
+
):
|
44 |
+
super().__init__()
|
45 |
+
self.positional_embedding = nn.Parameter(th.randn(embed_dim, spacial_dim ** 2 + 1) / embed_dim ** 0.5)
|
46 |
+
self.qkv_proj = conv_nd(1, embed_dim, 3 * embed_dim, 1)
|
47 |
+
self.c_proj = conv_nd(1, embed_dim, output_dim or embed_dim, 1)
|
48 |
+
self.num_heads = embed_dim // num_heads_channels
|
49 |
+
self.attention = QKVAttention(self.num_heads)
|
50 |
+
|
51 |
+
def forward(self, x):
|
52 |
+
b, c, *_spatial = x.shape
|
53 |
+
x = x.reshape(b, c, -1) # NC(HW)
|
54 |
+
x = th.cat([x.mean(dim=-1, keepdim=True), x], dim=-1) # NC(HW+1)
|
55 |
+
x = x + self.positional_embedding[None, :, :].to(x.dtype) # NC(HW+1)
|
56 |
+
x = self.qkv_proj(x)
|
57 |
+
x = self.attention(x)
|
58 |
+
x = self.c_proj(x)
|
59 |
+
return x[:, :, 0]
|
60 |
+
|
61 |
+
|
62 |
+
class TimestepBlock(nn.Module):
|
63 |
+
"""
|
64 |
+
Any module where forward() takes timestep embeddings as a second argument.
|
65 |
+
"""
|
66 |
+
|
67 |
+
@abstractmethod
|
68 |
+
def forward(self, x, emb):
|
69 |
+
"""
|
70 |
+
Apply the module to `x` given `emb` timestep embeddings.
|
71 |
+
"""
|
72 |
+
|
73 |
+
|
74 |
+
class TimestepEmbedSequential(nn.Sequential, TimestepBlock):
|
75 |
+
"""
|
76 |
+
A sequential module that passes timestep embeddings to the children that
|
77 |
+
support it as an extra input.
|
78 |
+
"""
|
79 |
+
|
80 |
+
def forward(self, x, emb, context=None, attn_dict=None, layer_type=None):
|
81 |
+
for layer in self:
|
82 |
+
if isinstance(layer, TimestepBlock):
|
83 |
+
x = layer(x, emb)
|
84 |
+
elif isinstance(layer, SpatialTransformer):
|
85 |
+
x = layer(x, context, attn_dict=attn_dict, layer_type=layer_type)
|
86 |
+
else:
|
87 |
+
x = layer(x)
|
88 |
+
return x
|
89 |
+
|
90 |
+
|
91 |
+
class Upsample(nn.Module):
|
92 |
+
"""
|
93 |
+
An upsampling layer with an optional convolution.
|
94 |
+
:param channels: channels in the inputs and outputs.
|
95 |
+
:param use_conv: a bool determining if a convolution is applied.
|
96 |
+
:param dims: determines if the signal is 1D, 2D, or 3D. If 3D, then
|
97 |
+
upsampling occurs in the inner-two dimensions.
|
98 |
+
"""
|
99 |
+
|
100 |
+
def __init__(self, channels, use_conv, dims=2, out_channels=None, padding=1):
|
101 |
+
super().__init__()
|
102 |
+
self.channels = channels
|
103 |
+
self.out_channels = out_channels or channels
|
104 |
+
self.use_conv = use_conv
|
105 |
+
self.dims = dims
|
106 |
+
if use_conv:
|
107 |
+
self.conv = conv_nd(dims, self.channels, self.out_channels, 3, padding=padding)
|
108 |
+
|
109 |
+
def forward(self, x):
|
110 |
+
assert x.shape[1] == self.channels
|
111 |
+
if self.dims == 3:
|
112 |
+
x = F.interpolate(
|
113 |
+
x, (x.shape[2], x.shape[3] * 2, x.shape[4] * 2), mode="nearest"
|
114 |
+
)
|
115 |
+
else:
|
116 |
+
x = F.interpolate(x, scale_factor=2, mode="nearest")
|
117 |
+
if self.use_conv:
|
118 |
+
x = self.conv(x)
|
119 |
+
return x
|
120 |
+
|
121 |
+
class TransposedUpsample(nn.Module):
|
122 |
+
'Learned 2x upsampling without padding'
|
123 |
+
def __init__(self, channels, out_channels=None, ks=5):
|
124 |
+
super().__init__()
|
125 |
+
self.channels = channels
|
126 |
+
self.out_channels = out_channels or channels
|
127 |
+
|
128 |
+
self.up = nn.ConvTranspose2d(self.channels,self.out_channels,kernel_size=ks,stride=2)
|
129 |
+
|
130 |
+
def forward(self,x):
|
131 |
+
return self.up(x)
|
132 |
+
|
133 |
+
|
134 |
+
class Downsample(nn.Module):
|
135 |
+
"""
|
136 |
+
A downsampling layer with an optional convolution.
|
137 |
+
:param channels: channels in the inputs and outputs.
|
138 |
+
:param use_conv: a bool determining if a convolution is applied.
|
139 |
+
:param dims: determines if the signal is 1D, 2D, or 3D. If 3D, then
|
140 |
+
downsampling occurs in the inner-two dimensions.
|
141 |
+
"""
|
142 |
+
|
143 |
+
def __init__(self, channels, use_conv, dims=2, out_channels=None,padding=1):
|
144 |
+
super().__init__()
|
145 |
+
self.channels = channels
|
146 |
+
self.out_channels = out_channels or channels
|
147 |
+
self.use_conv = use_conv
|
148 |
+
self.dims = dims
|
149 |
+
stride = 2 if dims != 3 else (1, 2, 2)
|
150 |
+
if use_conv:
|
151 |
+
self.op = conv_nd(
|
152 |
+
dims, self.channels, self.out_channels, 3, stride=stride, padding=padding
|
153 |
+
)
|
154 |
+
else:
|
155 |
+
assert self.channels == self.out_channels
|
156 |
+
self.op = avg_pool_nd(dims, kernel_size=stride, stride=stride)
|
157 |
+
|
158 |
+
def forward(self, x):
|
159 |
+
assert x.shape[1] == self.channels
|
160 |
+
return self.op(x)
|
161 |
+
|
162 |
+
|
163 |
+
class ResBlock(TimestepBlock):
|
164 |
+
"""
|
165 |
+
A residual block that can optionally change the number of channels.
|
166 |
+
:param channels: the number of input channels.
|
167 |
+
:param emb_channels: the number of timestep embedding channels.
|
168 |
+
:param dropout: the rate of dropout.
|
169 |
+
:param out_channels: if specified, the number of out channels.
|
170 |
+
:param use_conv: if True and out_channels is specified, use a spatial
|
171 |
+
convolution instead of a smaller 1x1 convolution to change the
|
172 |
+
channels in the skip connection.
|
173 |
+
:param dims: determines if the signal is 1D, 2D, or 3D.
|
174 |
+
:param use_checkpoint: if True, use gradient checkpointing on this module.
|
175 |
+
:param up: if True, use this block for upsampling.
|
176 |
+
:param down: if True, use this block for downsampling.
|
177 |
+
"""
|
178 |
+
|
179 |
+
def __init__(
|
180 |
+
self,
|
181 |
+
channels,
|
182 |
+
emb_channels,
|
183 |
+
dropout,
|
184 |
+
out_channels=None,
|
185 |
+
use_conv=False,
|
186 |
+
use_scale_shift_norm=False,
|
187 |
+
dims=2,
|
188 |
+
use_checkpoint=False,
|
189 |
+
up=False,
|
190 |
+
down=False,
|
191 |
+
):
|
192 |
+
super().__init__()
|
193 |
+
self.channels = channels
|
194 |
+
self.emb_channels = emb_channels
|
195 |
+
self.dropout = dropout
|
196 |
+
self.out_channels = out_channels or channels
|
197 |
+
self.use_conv = use_conv
|
198 |
+
self.use_checkpoint = use_checkpoint
|
199 |
+
self.use_scale_shift_norm = use_scale_shift_norm
|
200 |
+
|
201 |
+
self.in_layers = nn.Sequential(
|
202 |
+
normalization(channels),
|
203 |
+
nn.SiLU(),
|
204 |
+
conv_nd(dims, channels, self.out_channels, 3, padding=1),
|
205 |
+
)
|
206 |
+
|
207 |
+
self.updown = up or down
|
208 |
+
|
209 |
+
if up:
|
210 |
+
self.h_upd = Upsample(channels, False, dims)
|
211 |
+
self.x_upd = Upsample(channels, False, dims)
|
212 |
+
elif down:
|
213 |
+
self.h_upd = Downsample(channels, False, dims)
|
214 |
+
self.x_upd = Downsample(channels, False, dims)
|
215 |
+
else:
|
216 |
+
self.h_upd = self.x_upd = nn.Identity()
|
217 |
+
|
218 |
+
self.emb_layers = nn.Sequential(
|
219 |
+
nn.SiLU(),
|
220 |
+
linear(
|
221 |
+
emb_channels,
|
222 |
+
2 * self.out_channels if use_scale_shift_norm else self.out_channels,
|
223 |
+
),
|
224 |
+
)
|
225 |
+
self.out_layers = nn.Sequential(
|
226 |
+
normalization(self.out_channels),
|
227 |
+
nn.SiLU(),
|
228 |
+
nn.Dropout(p=dropout),
|
229 |
+
zero_module(
|
230 |
+
conv_nd(dims, self.out_channels, self.out_channels, 3, padding=1)
|
231 |
+
),
|
232 |
+
)
|
233 |
+
|
234 |
+
if self.out_channels == channels:
|
235 |
+
self.skip_connection = nn.Identity()
|
236 |
+
elif use_conv:
|
237 |
+
self.skip_connection = conv_nd(
|
238 |
+
dims, channels, self.out_channels, 3, padding=1
|
239 |
+
)
|
240 |
+
else:
|
241 |
+
self.skip_connection = conv_nd(dims, channels, self.out_channels, 1)
|
242 |
+
|
243 |
+
def forward(self, x, emb):
|
244 |
+
"""
|
245 |
+
Apply the block to a Tensor, conditioned on a timestep embedding.
|
246 |
+
:param x: an [N x C x ...] Tensor of features.
|
247 |
+
:param emb: an [N x emb_channels] Tensor of timestep embeddings.
|
248 |
+
:return: an [N x C x ...] Tensor of outputs.
|
249 |
+
"""
|
250 |
+
return checkpoint(
|
251 |
+
self._forward, (x, emb), self.parameters(), self.use_checkpoint
|
252 |
+
)
|
253 |
+
|
254 |
+
|
255 |
+
def _forward(self, x, emb):
|
256 |
+
if self.updown:
|
257 |
+
in_rest, in_conv = self.in_layers[:-1], self.in_layers[-1]
|
258 |
+
h = in_rest(x)
|
259 |
+
h = self.h_upd(h)
|
260 |
+
x = self.x_upd(x)
|
261 |
+
h = in_conv(h)
|
262 |
+
else:
|
263 |
+
h = self.in_layers(x)
|
264 |
+
emb_out = self.emb_layers(emb).type(h.dtype)
|
265 |
+
while len(emb_out.shape) < len(h.shape):
|
266 |
+
emb_out = emb_out[..., None]
|
267 |
+
if self.use_scale_shift_norm:
|
268 |
+
out_norm, out_rest = self.out_layers[0], self.out_layers[1:]
|
269 |
+
scale, shift = th.chunk(emb_out, 2, dim=1)
|
270 |
+
h = out_norm(h) * (1 + scale) + shift
|
271 |
+
h = out_rest(h)
|
272 |
+
else:
|
273 |
+
h = h + emb_out
|
274 |
+
h = self.out_layers(h)
|
275 |
+
return self.skip_connection(x) + h
|
276 |
+
|
277 |
+
|
278 |
+
class AttentionBlock(nn.Module):
|
279 |
+
"""
|
280 |
+
An attention block that allows spatial positions to attend to each other.
|
281 |
+
Originally ported from here, but adapted to the N-d case.
|
282 |
+
https://github.com/hojonathanho/diffusion/blob/1e0dceb3b3495bbe19116a5e1b3596cd0706c543/diffusion_tf/models/unet.py#L66.
|
283 |
+
"""
|
284 |
+
|
285 |
+
def __init__(
|
286 |
+
self,
|
287 |
+
channels,
|
288 |
+
num_heads=1,
|
289 |
+
num_head_channels=-1,
|
290 |
+
use_checkpoint=False,
|
291 |
+
use_new_attention_order=False,
|
292 |
+
):
|
293 |
+
super().__init__()
|
294 |
+
self.channels = channels
|
295 |
+
if num_head_channels == -1:
|
296 |
+
self.num_heads = num_heads
|
297 |
+
else:
|
298 |
+
assert (
|
299 |
+
channels % num_head_channels == 0
|
300 |
+
), f"q,k,v channels {channels} is not divisible by num_head_channels {num_head_channels}"
|
301 |
+
self.num_heads = channels // num_head_channels
|
302 |
+
self.use_checkpoint = use_checkpoint
|
303 |
+
self.norm = normalization(channels)
|
304 |
+
self.qkv = conv_nd(1, channels, channels * 3, 1)
|
305 |
+
if use_new_attention_order:
|
306 |
+
# split qkv before split heads
|
307 |
+
self.attention = QKVAttention(self.num_heads)
|
308 |
+
else:
|
309 |
+
# split heads before split qkv
|
310 |
+
self.attention = QKVAttentionLegacy(self.num_heads)
|
311 |
+
|
312 |
+
self.proj_out = zero_module(conv_nd(1, channels, channels, 1))
|
313 |
+
|
314 |
+
def forward(self, x):
|
315 |
+
return checkpoint(self._forward, (x,), self.parameters(), True) # TODO: check checkpoint usage, is True # TODO: fix the .half call!!!
|
316 |
+
#return pt_checkpoint(self._forward, x) # pytorch
|
317 |
+
|
318 |
+
def _forward(self, x):
|
319 |
+
b, c, *spatial = x.shape
|
320 |
+
x = x.reshape(b, c, -1)
|
321 |
+
qkv = self.qkv(self.norm(x))
|
322 |
+
h = self.attention(qkv)
|
323 |
+
h = self.proj_out(h)
|
324 |
+
return (x + h).reshape(b, c, *spatial)
|
325 |
+
|
326 |
+
|
327 |
+
def count_flops_attn(model, _x, y):
|
328 |
+
"""
|
329 |
+
A counter for the `thop` package to count the operations in an
|
330 |
+
attention operation.
|
331 |
+
Meant to be used like:
|
332 |
+
macs, params = thop.profile(
|
333 |
+
model,
|
334 |
+
inputs=(inputs, timestamps),
|
335 |
+
custom_ops={QKVAttention: QKVAttention.count_flops},
|
336 |
+
)
|
337 |
+
"""
|
338 |
+
b, c, *spatial = y[0].shape
|
339 |
+
num_spatial = int(np.prod(spatial))
|
340 |
+
# We perform two matmuls with the same number of ops.
|
341 |
+
# The first computes the weight matrix, the second computes
|
342 |
+
# the combination of the value vectors.
|
343 |
+
matmul_ops = 2 * b * (num_spatial ** 2) * c
|
344 |
+
model.total_ops += th.DoubleTensor([matmul_ops])
|
345 |
+
|
346 |
+
|
347 |
+
class QKVAttentionLegacy(nn.Module):
|
348 |
+
"""
|
349 |
+
A module which performs QKV attention. Matches legacy QKVAttention + input/ouput heads shaping
|
350 |
+
"""
|
351 |
+
|
352 |
+
def __init__(self, n_heads):
|
353 |
+
super().__init__()
|
354 |
+
self.n_heads = n_heads
|
355 |
+
|
356 |
+
def forward(self, qkv):
|
357 |
+
"""
|
358 |
+
Apply QKV attention.
|
359 |
+
:param qkv: an [N x (H * 3 * C) x T] tensor of Qs, Ks, and Vs.
|
360 |
+
:return: an [N x (H * C) x T] tensor after attention.
|
361 |
+
"""
|
362 |
+
bs, width, length = qkv.shape
|
363 |
+
assert width % (3 * self.n_heads) == 0
|
364 |
+
ch = width // (3 * self.n_heads)
|
365 |
+
q, k, v = qkv.reshape(bs * self.n_heads, ch * 3, length).split(ch, dim=1)
|
366 |
+
scale = 1 / math.sqrt(math.sqrt(ch))
|
367 |
+
weight = th.einsum(
|
368 |
+
"bct,bcs->bts", q * scale, k * scale
|
369 |
+
) # More stable with f16 than dividing afterwards
|
370 |
+
weight = th.softmax(weight.float(), dim=-1).type(weight.dtype)
|
371 |
+
a = th.einsum("bts,bcs->bct", weight, v)
|
372 |
+
return a.reshape(bs, -1, length)
|
373 |
+
|
374 |
+
@staticmethod
|
375 |
+
def count_flops(model, _x, y):
|
376 |
+
return count_flops_attn(model, _x, y)
|
377 |
+
|
378 |
+
|
379 |
+
class QKVAttention(nn.Module):
|
380 |
+
"""
|
381 |
+
A module which performs QKV attention and splits in a different order.
|
382 |
+
"""
|
383 |
+
|
384 |
+
def __init__(self, n_heads):
|
385 |
+
super().__init__()
|
386 |
+
self.n_heads = n_heads
|
387 |
+
|
388 |
+
def forward(self, qkv):
|
389 |
+
"""
|
390 |
+
Apply QKV attention.
|
391 |
+
:param qkv: an [N x (3 * H * C) x T] tensor of Qs, Ks, and Vs.
|
392 |
+
:return: an [N x (H * C) x T] tensor after attention.
|
393 |
+
"""
|
394 |
+
bs, width, length = qkv.shape
|
395 |
+
assert width % (3 * self.n_heads) == 0
|
396 |
+
ch = width // (3 * self.n_heads)
|
397 |
+
q, k, v = qkv.chunk(3, dim=1)
|
398 |
+
scale = 1 / math.sqrt(math.sqrt(ch))
|
399 |
+
weight = th.einsum(
|
400 |
+
"bct,bcs->bts",
|
401 |
+
(q * scale).view(bs * self.n_heads, ch, length),
|
402 |
+
(k * scale).view(bs * self.n_heads, ch, length),
|
403 |
+
) # More stable with f16 than dividing afterwards
|
404 |
+
weight = th.softmax(weight.float(), dim=-1).type(weight.dtype)
|
405 |
+
a = th.einsum("bts,bcs->bct", weight, v.reshape(bs * self.n_heads, ch, length))
|
406 |
+
return a.reshape(bs, -1, length)
|
407 |
+
|
408 |
+
@staticmethod
|
409 |
+
def count_flops(model, _x, y):
|
410 |
+
return count_flops_attn(model, _x, y)
|
411 |
+
|
412 |
+
|
413 |
+
class UNetModel(nn.Module):
|
414 |
+
"""
|
415 |
+
The full UNet model with attention and timestep embedding.
|
416 |
+
:param in_channels: channels in the input Tensor.
|
417 |
+
:param model_channels: base channel count for the model.
|
418 |
+
:param out_channels: channels in the output Tensor.
|
419 |
+
:param num_res_blocks: number of residual blocks per downsample.
|
420 |
+
:param attention_resolutions: a collection of downsample rates at which
|
421 |
+
attention will take place. May be a set, list, or tuple.
|
422 |
+
For example, if this contains 4, then at 4x downsampling, attention
|
423 |
+
will be used.
|
424 |
+
:param dropout: the dropout probability.
|
425 |
+
:param channel_mult: channel multiplier for each level of the UNet.
|
426 |
+
:param conv_resample: if True, use learned convolutions for upsampling and
|
427 |
+
downsampling.
|
428 |
+
:param dims: determines if the signal is 1D, 2D, or 3D.
|
429 |
+
:param num_classes: if specified (as an int), then this model will be
|
430 |
+
class-conditional with `num_classes` classes.
|
431 |
+
:param use_checkpoint: use gradient checkpointing to reduce memory usage.
|
432 |
+
:param num_heads: the number of attention heads in each attention layer.
|
433 |
+
:param num_heads_channels: if specified, ignore num_heads and instead use
|
434 |
+
a fixed channel width per attention head.
|
435 |
+
:param num_heads_upsample: works with num_heads to set a different number
|
436 |
+
of heads for upsampling. Deprecated.
|
437 |
+
:param use_scale_shift_norm: use a FiLM-like conditioning mechanism.
|
438 |
+
:param resblock_updown: use residual blocks for up/downsampling.
|
439 |
+
:param use_new_attention_order: use a different attention pattern for potentially
|
440 |
+
increased efficiency.
|
441 |
+
"""
|
442 |
+
|
443 |
+
def __init__(
|
444 |
+
self,
|
445 |
+
image_size,
|
446 |
+
in_channels,
|
447 |
+
model_channels,
|
448 |
+
out_channels,
|
449 |
+
num_res_blocks,
|
450 |
+
attention_resolutions,
|
451 |
+
dropout=0,
|
452 |
+
channel_mult=(1, 2, 4, 8),
|
453 |
+
conv_resample=True,
|
454 |
+
dims=2,
|
455 |
+
num_classes=None,
|
456 |
+
use_checkpoint=False,
|
457 |
+
use_fp16=False,
|
458 |
+
num_heads=-1,
|
459 |
+
num_head_channels=-1,
|
460 |
+
num_heads_upsample=-1,
|
461 |
+
use_scale_shift_norm=False,
|
462 |
+
resblock_updown=False,
|
463 |
+
use_new_attention_order=False,
|
464 |
+
use_spatial_transformer=False, # custom transformer support
|
465 |
+
transformer_depth=1, # custom transformer support
|
466 |
+
context_dim=None, # custom transformer support
|
467 |
+
n_embed=None, # custom support for prediction of discrete ids into codebook of first stage vq model
|
468 |
+
legacy=True
|
469 |
+
):
|
470 |
+
super().__init__()
|
471 |
+
if use_spatial_transformer:
|
472 |
+
assert context_dim is not None, 'Fool!! You forgot to include the dimension of your cross-attention conditioning...'
|
473 |
+
|
474 |
+
if context_dim is not None:
|
475 |
+
assert use_spatial_transformer, 'Fool!! You forgot to use the spatial transformer for your cross-attention conditioning...'
|
476 |
+
from omegaconf.listconfig import ListConfig
|
477 |
+
if type(context_dim) == ListConfig:
|
478 |
+
context_dim = list(context_dim)
|
479 |
+
|
480 |
+
if num_heads_upsample == -1:
|
481 |
+
num_heads_upsample = num_heads
|
482 |
+
|
483 |
+
if num_heads == -1:
|
484 |
+
assert num_head_channels != -1, 'Either num_heads or num_head_channels has to be set'
|
485 |
+
|
486 |
+
if num_head_channels == -1:
|
487 |
+
assert num_heads != -1, 'Either num_heads or num_head_channels has to be set'
|
488 |
+
|
489 |
+
self.image_size = image_size
|
490 |
+
self.in_channels = in_channels
|
491 |
+
self.model_channels = model_channels
|
492 |
+
self.out_channels = out_channels
|
493 |
+
self.num_res_blocks = num_res_blocks
|
494 |
+
self.attention_resolutions = attention_resolutions
|
495 |
+
self.dropout = dropout
|
496 |
+
self.channel_mult = channel_mult
|
497 |
+
self.conv_resample = conv_resample
|
498 |
+
self.num_classes = num_classes
|
499 |
+
self.use_checkpoint = use_checkpoint
|
500 |
+
self.dtype = th.float16 if use_fp16 else th.float32
|
501 |
+
self.num_heads = num_heads
|
502 |
+
self.num_head_channels = num_head_channels
|
503 |
+
self.num_heads_upsample = num_heads_upsample
|
504 |
+
self.predict_codebook_ids = n_embed is not None
|
505 |
+
|
506 |
+
time_embed_dim = model_channels * 4
|
507 |
+
self.time_embed = nn.Sequential(
|
508 |
+
linear(model_channels, time_embed_dim),
|
509 |
+
nn.SiLU(),
|
510 |
+
linear(time_embed_dim, time_embed_dim),
|
511 |
+
)
|
512 |
+
|
513 |
+
if self.num_classes is not None:
|
514 |
+
self.label_emb = nn.Embedding(num_classes, time_embed_dim)
|
515 |
+
|
516 |
+
self.input_blocks = nn.ModuleList(
|
517 |
+
[
|
518 |
+
TimestepEmbedSequential(
|
519 |
+
conv_nd(dims, in_channels, model_channels, 3, padding=1)
|
520 |
+
)
|
521 |
+
]
|
522 |
+
)
|
523 |
+
self._feature_size = model_channels
|
524 |
+
input_block_chans = [model_channels]
|
525 |
+
ch = model_channels
|
526 |
+
ds = 1
|
527 |
+
for level, mult in enumerate(channel_mult):
|
528 |
+
for _ in range(num_res_blocks):
|
529 |
+
layers = [
|
530 |
+
ResBlock(
|
531 |
+
ch,
|
532 |
+
time_embed_dim,
|
533 |
+
dropout,
|
534 |
+
out_channels=mult * model_channels,
|
535 |
+
dims=dims,
|
536 |
+
use_checkpoint=use_checkpoint,
|
537 |
+
use_scale_shift_norm=use_scale_shift_norm,
|
538 |
+
)
|
539 |
+
]
|
540 |
+
ch = mult * model_channels
|
541 |
+
if ds in attention_resolutions:
|
542 |
+
if num_head_channels == -1:
|
543 |
+
dim_head = ch // num_heads
|
544 |
+
else:
|
545 |
+
num_heads = ch // num_head_channels
|
546 |
+
dim_head = num_head_channels
|
547 |
+
if legacy:
|
548 |
+
#num_heads = 1
|
549 |
+
dim_head = ch // num_heads if use_spatial_transformer else num_head_channels
|
550 |
+
layers.append(
|
551 |
+
AttentionBlock(
|
552 |
+
ch,
|
553 |
+
use_checkpoint=use_checkpoint,
|
554 |
+
num_heads=num_heads,
|
555 |
+
num_head_channels=dim_head,
|
556 |
+
use_new_attention_order=use_new_attention_order,
|
557 |
+
) if not use_spatial_transformer else SpatialTransformer(
|
558 |
+
ch, num_heads, dim_head, depth=transformer_depth, context_dim=context_dim
|
559 |
+
)
|
560 |
+
)
|
561 |
+
self.input_blocks.append(TimestepEmbedSequential(*layers))
|
562 |
+
self._feature_size += ch
|
563 |
+
input_block_chans.append(ch)
|
564 |
+
if level != len(channel_mult) - 1:
|
565 |
+
out_ch = ch
|
566 |
+
self.input_blocks.append(
|
567 |
+
TimestepEmbedSequential(
|
568 |
+
ResBlock(
|
569 |
+
ch,
|
570 |
+
time_embed_dim,
|
571 |
+
dropout,
|
572 |
+
out_channels=out_ch,
|
573 |
+
dims=dims,
|
574 |
+
use_checkpoint=use_checkpoint,
|
575 |
+
use_scale_shift_norm=use_scale_shift_norm,
|
576 |
+
down=True,
|
577 |
+
)
|
578 |
+
if resblock_updown
|
579 |
+
else Downsample(
|
580 |
+
ch, conv_resample, dims=dims, out_channels=out_ch
|
581 |
+
)
|
582 |
+
)
|
583 |
+
)
|
584 |
+
ch = out_ch
|
585 |
+
input_block_chans.append(ch)
|
586 |
+
ds *= 2
|
587 |
+
self._feature_size += ch
|
588 |
+
|
589 |
+
if num_head_channels == -1:
|
590 |
+
dim_head = ch // num_heads
|
591 |
+
else:
|
592 |
+
num_heads = ch // num_head_channels
|
593 |
+
dim_head = num_head_channels
|
594 |
+
if legacy:
|
595 |
+
#num_heads = 1
|
596 |
+
dim_head = ch // num_heads if use_spatial_transformer else num_head_channels
|
597 |
+
self.middle_block = TimestepEmbedSequential(
|
598 |
+
ResBlock(
|
599 |
+
ch,
|
600 |
+
time_embed_dim,
|
601 |
+
dropout,
|
602 |
+
dims=dims,
|
603 |
+
use_checkpoint=use_checkpoint,
|
604 |
+
use_scale_shift_norm=use_scale_shift_norm,
|
605 |
+
),
|
606 |
+
AttentionBlock(
|
607 |
+
ch,
|
608 |
+
use_checkpoint=use_checkpoint,
|
609 |
+
num_heads=num_heads,
|
610 |
+
num_head_channels=dim_head,
|
611 |
+
use_new_attention_order=use_new_attention_order,
|
612 |
+
) if not use_spatial_transformer else SpatialTransformer(
|
613 |
+
ch, num_heads, dim_head, depth=transformer_depth, context_dim=context_dim
|
614 |
+
),
|
615 |
+
ResBlock(
|
616 |
+
ch,
|
617 |
+
time_embed_dim,
|
618 |
+
dropout,
|
619 |
+
dims=dims,
|
620 |
+
use_checkpoint=use_checkpoint,
|
621 |
+
use_scale_shift_norm=use_scale_shift_norm,
|
622 |
+
),
|
623 |
+
)
|
624 |
+
self._feature_size += ch
|
625 |
+
|
626 |
+
self.output_blocks = nn.ModuleList([])
|
627 |
+
for level, mult in list(enumerate(channel_mult))[::-1]:
|
628 |
+
for i in range(num_res_blocks + 1):
|
629 |
+
ich = input_block_chans.pop()
|
630 |
+
layers = [
|
631 |
+
ResBlock(
|
632 |
+
ch + ich,
|
633 |
+
time_embed_dim,
|
634 |
+
dropout,
|
635 |
+
out_channels=model_channels * mult,
|
636 |
+
dims=dims,
|
637 |
+
use_checkpoint=use_checkpoint,
|
638 |
+
use_scale_shift_norm=use_scale_shift_norm,
|
639 |
+
)
|
640 |
+
]
|
641 |
+
ch = model_channels * mult
|
642 |
+
if ds in attention_resolutions:
|
643 |
+
if num_head_channels == -1:
|
644 |
+
dim_head = ch // num_heads
|
645 |
+
else:
|
646 |
+
num_heads = ch // num_head_channels
|
647 |
+
dim_head = num_head_channels
|
648 |
+
if legacy:
|
649 |
+
#num_heads = 1
|
650 |
+
dim_head = ch // num_heads if use_spatial_transformer else num_head_channels
|
651 |
+
layers.append(
|
652 |
+
AttentionBlock(
|
653 |
+
ch,
|
654 |
+
use_checkpoint=use_checkpoint,
|
655 |
+
num_heads=num_heads_upsample,
|
656 |
+
num_head_channels=dim_head,
|
657 |
+
use_new_attention_order=use_new_attention_order,
|
658 |
+
) if not use_spatial_transformer else SpatialTransformer(
|
659 |
+
ch, num_heads, dim_head, depth=transformer_depth, context_dim=context_dim
|
660 |
+
)
|
661 |
+
)
|
662 |
+
if level and i == num_res_blocks:
|
663 |
+
out_ch = ch
|
664 |
+
layers.append(
|
665 |
+
ResBlock(
|
666 |
+
ch,
|
667 |
+
time_embed_dim,
|
668 |
+
dropout,
|
669 |
+
out_channels=out_ch,
|
670 |
+
dims=dims,
|
671 |
+
use_checkpoint=use_checkpoint,
|
672 |
+
use_scale_shift_norm=use_scale_shift_norm,
|
673 |
+
up=True,
|
674 |
+
)
|
675 |
+
if resblock_updown
|
676 |
+
else Upsample(ch, conv_resample, dims=dims, out_channels=out_ch)
|
677 |
+
)
|
678 |
+
ds //= 2
|
679 |
+
self.output_blocks.append(TimestepEmbedSequential(*layers))
|
680 |
+
self._feature_size += ch
|
681 |
+
|
682 |
+
self.out = nn.Sequential(
|
683 |
+
normalization(ch),
|
684 |
+
nn.SiLU(),
|
685 |
+
zero_module(conv_nd(dims, model_channels, out_channels, 3, padding=1)),
|
686 |
+
)
|
687 |
+
if self.predict_codebook_ids:
|
688 |
+
self.id_predictor = nn.Sequential(
|
689 |
+
normalization(ch),
|
690 |
+
conv_nd(dims, model_channels, n_embed, 1),
|
691 |
+
#nn.LogSoftmax(dim=1) # change to cross_entropy and produce non-normalized logits
|
692 |
+
)
|
693 |
+
|
694 |
+
def convert_to_fp16(self):
|
695 |
+
"""
|
696 |
+
Convert the torso of the model to float16.
|
697 |
+
"""
|
698 |
+
self.input_blocks.apply(convert_module_to_f16)
|
699 |
+
self.middle_block.apply(convert_module_to_f16)
|
700 |
+
self.output_blocks.apply(convert_module_to_f16)
|
701 |
+
|
702 |
+
def convert_to_fp32(self):
|
703 |
+
"""
|
704 |
+
Convert the torso of the model to float32.
|
705 |
+
"""
|
706 |
+
self.input_blocks.apply(convert_module_to_f32)
|
707 |
+
self.middle_block.apply(convert_module_to_f32)
|
708 |
+
self.output_blocks.apply(convert_module_to_f32)
|
709 |
+
|
710 |
+
def forward(self, x, timesteps=None, context=None, y=None, attn_dict=None, **kwargs):
|
711 |
+
# attn_dict attribute is used to return the attention valus for visualization
|
712 |
+
"""
|
713 |
+
Apply the model to an input batch.
|
714 |
+
:param x: an [N x C x ...] Tensor of inputs.
|
715 |
+
:param timesteps: a 1-D batch of timesteps.
|
716 |
+
:param context: conditioning plugged in via crossattn
|
717 |
+
:param y: an [N] Tensor of labels, if class-conditional.
|
718 |
+
:return: an [N x C x ...] Tensor of outputs.
|
719 |
+
"""
|
720 |
+
keep_attns = True if attn_dict is not None else False
|
721 |
+
|
722 |
+
assert (y is not None) == (
|
723 |
+
self.num_classes is not None
|
724 |
+
), "must specify y if and only if the model is class-conditional"
|
725 |
+
hs = []
|
726 |
+
t_emb = timestep_embedding(timesteps, self.model_channels, repeat_only=False)
|
727 |
+
emb = self.time_embed(t_emb)
|
728 |
+
|
729 |
+
if self.num_classes is not None:
|
730 |
+
assert y.shape == (x.shape[0],)
|
731 |
+
emb = emb + self.label_emb(y)
|
732 |
+
|
733 |
+
h = x.type(self.dtype)
|
734 |
+
for module in self.input_blocks:
|
735 |
+
if keep_attns:
|
736 |
+
h = module(h, emb, context, attn_dict=attn_dict, layer_type="down")
|
737 |
+
else:
|
738 |
+
h = module(h, emb, context)
|
739 |
+
hs.append(h)
|
740 |
+
if keep_attns:
|
741 |
+
h = self.middle_block(h, emb, context, attn_dict=attn_dict, layer_type="middle")
|
742 |
+
else:
|
743 |
+
h = self.middle_block(h, emb, context)
|
744 |
+
for module in self.output_blocks:
|
745 |
+
h = th.cat([h, hs.pop()], dim=1)
|
746 |
+
if keep_attns:
|
747 |
+
h = module(h, emb, context, attn_dict=attn_dict, layer_type="up")
|
748 |
+
else:
|
749 |
+
h = module(h, emb, context)
|
750 |
+
h = h.type(x.dtype)
|
751 |
+
if self.predict_codebook_ids:
|
752 |
+
return self.id_predictor(h)
|
753 |
+
else:
|
754 |
+
return self.out(h)
|
755 |
+
|
756 |
+
|
757 |
+
class EncoderUNetModel(nn.Module):
|
758 |
+
"""
|
759 |
+
The half UNet model with attention and timestep embedding.
|
760 |
+
For usage, see UNet.
|
761 |
+
"""
|
762 |
+
|
763 |
+
def __init__(
|
764 |
+
self,
|
765 |
+
image_size,
|
766 |
+
in_channels,
|
767 |
+
model_channels,
|
768 |
+
out_channels,
|
769 |
+
num_res_blocks,
|
770 |
+
attention_resolutions,
|
771 |
+
dropout=0,
|
772 |
+
channel_mult=(1, 2, 4, 8),
|
773 |
+
conv_resample=True,
|
774 |
+
dims=2,
|
775 |
+
use_checkpoint=False,
|
776 |
+
use_fp16=False,
|
777 |
+
num_heads=1,
|
778 |
+
num_head_channels=-1,
|
779 |
+
num_heads_upsample=-1,
|
780 |
+
use_scale_shift_norm=False,
|
781 |
+
resblock_updown=False,
|
782 |
+
use_new_attention_order=False,
|
783 |
+
pool="adaptive",
|
784 |
+
*args,
|
785 |
+
**kwargs
|
786 |
+
):
|
787 |
+
super().__init__()
|
788 |
+
|
789 |
+
if num_heads_upsample == -1:
|
790 |
+
num_heads_upsample = num_heads
|
791 |
+
|
792 |
+
self.in_channels = in_channels
|
793 |
+
self.model_channels = model_channels
|
794 |
+
self.out_channels = out_channels
|
795 |
+
self.num_res_blocks = num_res_blocks
|
796 |
+
self.attention_resolutions = attention_resolutions
|
797 |
+
self.dropout = dropout
|
798 |
+
self.channel_mult = channel_mult
|
799 |
+
self.conv_resample = conv_resample
|
800 |
+
self.use_checkpoint = use_checkpoint
|
801 |
+
self.dtype = th.float16 if use_fp16 else th.float32
|
802 |
+
self.num_heads = num_heads
|
803 |
+
self.num_head_channels = num_head_channels
|
804 |
+
self.num_heads_upsample = num_heads_upsample
|
805 |
+
|
806 |
+
time_embed_dim = model_channels * 4
|
807 |
+
self.time_embed = nn.Sequential(
|
808 |
+
linear(model_channels, time_embed_dim),
|
809 |
+
nn.SiLU(),
|
810 |
+
linear(time_embed_dim, time_embed_dim),
|
811 |
+
)
|
812 |
+
|
813 |
+
self.input_blocks = nn.ModuleList(
|
814 |
+
[
|
815 |
+
TimestepEmbedSequential(
|
816 |
+
conv_nd(dims, in_channels, model_channels, 3, padding=1)
|
817 |
+
)
|
818 |
+
]
|
819 |
+
)
|
820 |
+
self._feature_size = model_channels
|
821 |
+
input_block_chans = [model_channels]
|
822 |
+
ch = model_channels
|
823 |
+
ds = 1
|
824 |
+
for level, mult in enumerate(channel_mult):
|
825 |
+
for _ in range(num_res_blocks):
|
826 |
+
layers = [
|
827 |
+
ResBlock(
|
828 |
+
ch,
|
829 |
+
time_embed_dim,
|
830 |
+
dropout,
|
831 |
+
out_channels=mult * model_channels,
|
832 |
+
dims=dims,
|
833 |
+
use_checkpoint=use_checkpoint,
|
834 |
+
use_scale_shift_norm=use_scale_shift_norm,
|
835 |
+
)
|
836 |
+
]
|
837 |
+
ch = mult * model_channels
|
838 |
+
if ds in attention_resolutions:
|
839 |
+
layers.append(
|
840 |
+
AttentionBlock(
|
841 |
+
ch,
|
842 |
+
use_checkpoint=use_checkpoint,
|
843 |
+
num_heads=num_heads,
|
844 |
+
num_head_channels=num_head_channels,
|
845 |
+
use_new_attention_order=use_new_attention_order,
|
846 |
+
)
|
847 |
+
)
|
848 |
+
self.input_blocks.append(TimestepEmbedSequential(*layers))
|
849 |
+
self._feature_size += ch
|
850 |
+
input_block_chans.append(ch)
|
851 |
+
if level != len(channel_mult) - 1:
|
852 |
+
out_ch = ch
|
853 |
+
self.input_blocks.append(
|
854 |
+
TimestepEmbedSequential(
|
855 |
+
ResBlock(
|
856 |
+
ch,
|
857 |
+
time_embed_dim,
|
858 |
+
dropout,
|
859 |
+
out_channels=out_ch,
|
860 |
+
dims=dims,
|
861 |
+
use_checkpoint=use_checkpoint,
|
862 |
+
use_scale_shift_norm=use_scale_shift_norm,
|
863 |
+
down=True,
|
864 |
+
)
|
865 |
+
if resblock_updown
|
866 |
+
else Downsample(
|
867 |
+
ch, conv_resample, dims=dims, out_channels=out_ch
|
868 |
+
)
|
869 |
+
)
|
870 |
+
)
|
871 |
+
ch = out_ch
|
872 |
+
input_block_chans.append(ch)
|
873 |
+
ds *= 2
|
874 |
+
self._feature_size += ch
|
875 |
+
|
876 |
+
self.middle_block = TimestepEmbedSequential(
|
877 |
+
ResBlock(
|
878 |
+
ch,
|
879 |
+
time_embed_dim,
|
880 |
+
dropout,
|
881 |
+
dims=dims,
|
882 |
+
use_checkpoint=use_checkpoint,
|
883 |
+
use_scale_shift_norm=use_scale_shift_norm,
|
884 |
+
),
|
885 |
+
AttentionBlock(
|
886 |
+
ch,
|
887 |
+
use_checkpoint=use_checkpoint,
|
888 |
+
num_heads=num_heads,
|
889 |
+
num_head_channels=num_head_channels,
|
890 |
+
use_new_attention_order=use_new_attention_order,
|
891 |
+
),
|
892 |
+
ResBlock(
|
893 |
+
ch,
|
894 |
+
time_embed_dim,
|
895 |
+
dropout,
|
896 |
+
dims=dims,
|
897 |
+
use_checkpoint=use_checkpoint,
|
898 |
+
use_scale_shift_norm=use_scale_shift_norm,
|
899 |
+
),
|
900 |
+
)
|
901 |
+
self._feature_size += ch
|
902 |
+
self.pool = pool
|
903 |
+
if pool == "adaptive":
|
904 |
+
self.out = nn.Sequential(
|
905 |
+
normalization(ch),
|
906 |
+
nn.SiLU(),
|
907 |
+
nn.AdaptiveAvgPool2d((1, 1)),
|
908 |
+
zero_module(conv_nd(dims, ch, out_channels, 1)),
|
909 |
+
nn.Flatten(),
|
910 |
+
)
|
911 |
+
elif pool == "attention":
|
912 |
+
assert num_head_channels != -1
|
913 |
+
self.out = nn.Sequential(
|
914 |
+
normalization(ch),
|
915 |
+
nn.SiLU(),
|
916 |
+
AttentionPool2d(
|
917 |
+
(image_size // ds), ch, num_head_channels, out_channels
|
918 |
+
),
|
919 |
+
)
|
920 |
+
elif pool == "spatial":
|
921 |
+
self.out = nn.Sequential(
|
922 |
+
nn.Linear(self._feature_size, 2048),
|
923 |
+
nn.ReLU(),
|
924 |
+
nn.Linear(2048, self.out_channels),
|
925 |
+
)
|
926 |
+
elif pool == "spatial_v2":
|
927 |
+
self.out = nn.Sequential(
|
928 |
+
nn.Linear(self._feature_size, 2048),
|
929 |
+
normalization(2048),
|
930 |
+
nn.SiLU(),
|
931 |
+
nn.Linear(2048, self.out_channels),
|
932 |
+
)
|
933 |
+
else:
|
934 |
+
raise NotImplementedError(f"Unexpected {pool} pooling")
|
935 |
+
|
936 |
+
def convert_to_fp16(self):
|
937 |
+
"""
|
938 |
+
Convert the torso of the model to float16.
|
939 |
+
"""
|
940 |
+
self.input_blocks.apply(convert_module_to_f16)
|
941 |
+
self.middle_block.apply(convert_module_to_f16)
|
942 |
+
|
943 |
+
def convert_to_fp32(self):
|
944 |
+
"""
|
945 |
+
Convert the torso of the model to float32.
|
946 |
+
"""
|
947 |
+
self.input_blocks.apply(convert_module_to_f32)
|
948 |
+
self.middle_block.apply(convert_module_to_f32)
|
949 |
+
|
950 |
+
def forward(self, x, timesteps):
|
951 |
+
"""
|
952 |
+
Apply the model to an input batch.
|
953 |
+
:param x: an [N x C x ...] Tensor of inputs.
|
954 |
+
:param timesteps: a 1-D batch of timesteps.
|
955 |
+
:return: an [N x K] Tensor of outputs.
|
956 |
+
"""
|
957 |
+
emb = self.time_embed(timestep_embedding(timesteps, self.model_channels))
|
958 |
+
|
959 |
+
results = []
|
960 |
+
h = x.type(self.dtype)
|
961 |
+
for module in self.input_blocks:
|
962 |
+
h = module(h, emb)
|
963 |
+
if self.pool.startswith("spatial"):
|
964 |
+
results.append(h.type(x.dtype).mean(dim=(2, 3)))
|
965 |
+
h = self.middle_block(h, emb)
|
966 |
+
if self.pool.startswith("spatial"):
|
967 |
+
results.append(h.type(x.dtype).mean(dim=(2, 3)))
|
968 |
+
h = th.cat(results, axis=-1)
|
969 |
+
return self.out(h)
|
970 |
+
else:
|
971 |
+
h = h.type(x.dtype)
|
972 |
+
return self.out(h)
|
973 |
+
|
ldm/modules/diffusionmodules/util.py
ADDED
@@ -0,0 +1,270 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# adopted from
|
2 |
+
# https://github.com/openai/improved-diffusion/blob/main/improved_diffusion/gaussian_diffusion.py
|
3 |
+
# and
|
4 |
+
# https://github.com/lucidrains/denoising-diffusion-pytorch/blob/7706bdfc6f527f58d33f84b7b522e61e6e3164b3/denoising_diffusion_pytorch/denoising_diffusion_pytorch.py
|
5 |
+
# and
|
6 |
+
# https://github.com/openai/guided-diffusion/blob/0ba878e517b276c45d1195eb29f6f5f72659a05b/guided_diffusion/nn.py
|
7 |
+
#
|
8 |
+
# thanks!
|
9 |
+
|
10 |
+
|
11 |
+
import os
|
12 |
+
import math
|
13 |
+
import torch
|
14 |
+
import torch.nn as nn
|
15 |
+
import numpy as np
|
16 |
+
from einops import repeat
|
17 |
+
|
18 |
+
from ldm.util import instantiate_from_config
|
19 |
+
|
20 |
+
|
21 |
+
def make_beta_schedule(schedule, n_timestep, linear_start=1e-4, linear_end=2e-2, cosine_s=8e-3):
|
22 |
+
if schedule == "linear":
|
23 |
+
betas = (
|
24 |
+
torch.linspace(linear_start ** 0.5, linear_end ** 0.5, n_timestep, dtype=torch.float64) ** 2
|
25 |
+
)
|
26 |
+
|
27 |
+
elif schedule == "cosine":
|
28 |
+
timesteps = (
|
29 |
+
torch.arange(n_timestep + 1, dtype=torch.float64) / n_timestep + cosine_s
|
30 |
+
)
|
31 |
+
alphas = timesteps / (1 + cosine_s) * np.pi / 2
|
32 |
+
alphas = torch.cos(alphas).pow(2)
|
33 |
+
alphas = alphas / alphas[0]
|
34 |
+
betas = 1 - alphas[1:] / alphas[:-1]
|
35 |
+
betas = np.clip(betas, a_min=0, a_max=0.999)
|
36 |
+
|
37 |
+
elif schedule == "sqrt_linear":
|
38 |
+
betas = torch.linspace(linear_start, linear_end, n_timestep, dtype=torch.float64)
|
39 |
+
elif schedule == "sqrt":
|
40 |
+
betas = torch.linspace(linear_start, linear_end, n_timestep, dtype=torch.float64) ** 0.5
|
41 |
+
else:
|
42 |
+
raise ValueError(f"schedule '{schedule}' unknown.")
|
43 |
+
return betas.numpy()
|
44 |
+
|
45 |
+
|
46 |
+
def make_ddim_timesteps(ddim_discr_method, num_ddim_timesteps, num_ddpm_timesteps, verbose=True):
|
47 |
+
if ddim_discr_method == 'uniform':
|
48 |
+
c = num_ddpm_timesteps // num_ddim_timesteps
|
49 |
+
ddim_timesteps = np.asarray(list(range(0, num_ddpm_timesteps, c)))
|
50 |
+
elif ddim_discr_method == 'quad':
|
51 |
+
ddim_timesteps = ((np.linspace(0, np.sqrt(num_ddpm_timesteps * .8), num_ddim_timesteps)) ** 2).astype(int)
|
52 |
+
else:
|
53 |
+
raise NotImplementedError(f'There is no ddim discretization method called "{ddim_discr_method}"')
|
54 |
+
|
55 |
+
# assert ddim_timesteps.shape[0] == num_ddim_timesteps
|
56 |
+
# add one to get the final alpha values right (the ones from first scale to data during sampling)
|
57 |
+
steps_out = ddim_timesteps + 1
|
58 |
+
if verbose:
|
59 |
+
print(f'Selected timesteps for ddim sampler: {steps_out}')
|
60 |
+
return steps_out
|
61 |
+
|
62 |
+
|
63 |
+
def make_ddim_sampling_parameters(alphacums, ddim_timesteps, eta, verbose=True):
|
64 |
+
# select alphas for computing the variance schedule
|
65 |
+
alphas = alphacums[ddim_timesteps]
|
66 |
+
alphas_prev = np.asarray([alphacums[0]] + alphacums[ddim_timesteps[:-1]].tolist())
|
67 |
+
|
68 |
+
# according the the formula provided in https://arxiv.org/abs/2010.02502
|
69 |
+
sigmas = eta * np.sqrt((1 - alphas_prev) / (1 - alphas) * (1 - alphas / alphas_prev))
|
70 |
+
if verbose:
|
71 |
+
print(f'Selected alphas for ddim sampler: a_t: {alphas}; a_(t-1): {alphas_prev}')
|
72 |
+
print(f'For the chosen value of eta, which is {eta}, '
|
73 |
+
f'this results in the following sigma_t schedule for ddim sampler {sigmas}')
|
74 |
+
return sigmas, alphas, alphas_prev
|
75 |
+
|
76 |
+
|
77 |
+
def betas_for_alpha_bar(num_diffusion_timesteps, alpha_bar, max_beta=0.999):
|
78 |
+
"""
|
79 |
+
Create a beta schedule that discretizes the given alpha_t_bar function,
|
80 |
+
which defines the cumulative product of (1-beta) over time from t = [0,1].
|
81 |
+
:param num_diffusion_timesteps: the number of betas to produce.
|
82 |
+
:param alpha_bar: a lambda that takes an argument t from 0 to 1 and
|
83 |
+
produces the cumulative product of (1-beta) up to that
|
84 |
+
part of the diffusion process.
|
85 |
+
:param max_beta: the maximum beta to use; use values lower than 1 to
|
86 |
+
prevent singularities.
|
87 |
+
"""
|
88 |
+
betas = []
|
89 |
+
for i in range(num_diffusion_timesteps):
|
90 |
+
t1 = i / num_diffusion_timesteps
|
91 |
+
t2 = (i + 1) / num_diffusion_timesteps
|
92 |
+
betas.append(min(1 - alpha_bar(t2) / alpha_bar(t1), max_beta))
|
93 |
+
return np.array(betas)
|
94 |
+
|
95 |
+
|
96 |
+
def extract_into_tensor(a, t, x_shape):
|
97 |
+
b, *_ = t.shape
|
98 |
+
out = a.gather(-1, t)
|
99 |
+
return out.reshape(b, *((1,) * (len(x_shape) - 1)))
|
100 |
+
|
101 |
+
|
102 |
+
def checkpoint(func, inputs, params, flag):
|
103 |
+
"""
|
104 |
+
Evaluate a function without caching intermediate activations, allowing for
|
105 |
+
reduced memory at the expense of extra compute in the backward pass.
|
106 |
+
:param func: the function to evaluate.
|
107 |
+
:param inputs: the argument sequence to pass to `func`.
|
108 |
+
:param params: a sequence of parameters `func` depends on but does not
|
109 |
+
explicitly take as arguments.
|
110 |
+
:param flag: if False, disable gradient checkpointing.
|
111 |
+
"""
|
112 |
+
if flag:
|
113 |
+
|
114 |
+
# For storing the attention maps, a dict object is sent to the forward function.
|
115 |
+
# Not to raise an error in below detach operation (backward function), dict and None objects are discarded.
|
116 |
+
inputs = [x for x in inputs if x is not isinstance(x, dict) and x is not None]
|
117 |
+
|
118 |
+
args = tuple(inputs) + tuple(params)
|
119 |
+
return CheckpointFunction.apply(func, len(inputs), *args)
|
120 |
+
else:
|
121 |
+
return func(*inputs)
|
122 |
+
|
123 |
+
class CheckpointFunction(torch.autograd.Function):
|
124 |
+
@staticmethod
|
125 |
+
def forward(ctx, run_function, length, *args):
|
126 |
+
ctx.run_function = run_function
|
127 |
+
ctx.input_tensors = list(args[:length])
|
128 |
+
ctx.input_params = list(args[length:])
|
129 |
+
with torch.no_grad():
|
130 |
+
output_tensors = ctx.run_function(*ctx.input_tensors)
|
131 |
+
return output_tensors
|
132 |
+
|
133 |
+
@staticmethod
|
134 |
+
def backward(ctx, *output_grads):
|
135 |
+
ctx.input_tensors = [x.detach().requires_grad_(True) for x in ctx.input_tensors]
|
136 |
+
with torch.enable_grad():
|
137 |
+
# Fixes a bug where the first op in run_function modifies the
|
138 |
+
# Tensor storage in place, which is not allowed for detach()'d
|
139 |
+
# Tensors.
|
140 |
+
shallow_copies = [x.view_as(x) for x in ctx.input_tensors]
|
141 |
+
output_tensors = ctx.run_function(*shallow_copies)
|
142 |
+
input_grads = torch.autograd.grad(
|
143 |
+
output_tensors,
|
144 |
+
ctx.input_tensors + ctx.input_params,
|
145 |
+
output_grads,
|
146 |
+
allow_unused=True,
|
147 |
+
)
|
148 |
+
del ctx.input_tensors
|
149 |
+
del ctx.input_params
|
150 |
+
del output_tensors
|
151 |
+
return (None, None) + input_grads
|
152 |
+
|
153 |
+
|
154 |
+
def timestep_embedding(timesteps, dim, max_period=10000, repeat_only=False):
|
155 |
+
"""
|
156 |
+
Create sinusoidal timestep embeddings.
|
157 |
+
:param timesteps: a 1-D Tensor of N indices, one per batch element.
|
158 |
+
These may be fractional.
|
159 |
+
:param dim: the dimension of the output.
|
160 |
+
:param max_period: controls the minimum frequency of the embeddings.
|
161 |
+
:return: an [N x dim] Tensor of positional embeddings.
|
162 |
+
"""
|
163 |
+
if not repeat_only:
|
164 |
+
half = dim // 2
|
165 |
+
freqs = torch.exp(
|
166 |
+
-math.log(max_period) * torch.arange(start=0, end=half, dtype=torch.float32) / half
|
167 |
+
).to(device=timesteps.device)
|
168 |
+
args = timesteps[:, None].float() * freqs[None]
|
169 |
+
embedding = torch.cat([torch.cos(args), torch.sin(args)], dim=-1)
|
170 |
+
if dim % 2:
|
171 |
+
embedding = torch.cat([embedding, torch.zeros_like(embedding[:, :1])], dim=-1)
|
172 |
+
else:
|
173 |
+
embedding = repeat(timesteps, 'b -> b d', d=dim)
|
174 |
+
return embedding
|
175 |
+
|
176 |
+
|
177 |
+
def zero_module(module):
|
178 |
+
"""
|
179 |
+
Zero out the parameters of a module and return it.
|
180 |
+
"""
|
181 |
+
for p in module.parameters():
|
182 |
+
p.detach().zero_()
|
183 |
+
return module
|
184 |
+
|
185 |
+
|
186 |
+
def scale_module(module, scale):
|
187 |
+
"""
|
188 |
+
Scale the parameters of a module and return it.
|
189 |
+
"""
|
190 |
+
for p in module.parameters():
|
191 |
+
p.detach().mul_(scale)
|
192 |
+
return module
|
193 |
+
|
194 |
+
|
195 |
+
def mean_flat(tensor):
|
196 |
+
"""
|
197 |
+
Take the mean over all non-batch dimensions.
|
198 |
+
"""
|
199 |
+
return tensor.mean(dim=list(range(1, len(tensor.shape))))
|
200 |
+
|
201 |
+
|
202 |
+
def normalization(channels):
|
203 |
+
"""
|
204 |
+
Make a standard normalization layer.
|
205 |
+
:param channels: number of input channels.
|
206 |
+
:return: an nn.Module for normalization.
|
207 |
+
"""
|
208 |
+
return GroupNorm32(32, channels)
|
209 |
+
|
210 |
+
|
211 |
+
# PyTorch 1.7 has SiLU, but we support PyTorch 1.5.
|
212 |
+
class SiLU(nn.Module):
|
213 |
+
def forward(self, x):
|
214 |
+
return x * torch.sigmoid(x)
|
215 |
+
|
216 |
+
|
217 |
+
class GroupNorm32(nn.GroupNorm):
|
218 |
+
def forward(self, x):
|
219 |
+
return super().forward(x.float()).type(x.dtype)
|
220 |
+
|
221 |
+
def conv_nd(dims, *args, **kwargs):
|
222 |
+
"""
|
223 |
+
Create a 1D, 2D, or 3D convolution module.
|
224 |
+
"""
|
225 |
+
if dims == 1:
|
226 |
+
return nn.Conv1d(*args, **kwargs)
|
227 |
+
elif dims == 2:
|
228 |
+
return nn.Conv2d(*args, **kwargs)
|
229 |
+
elif dims == 3:
|
230 |
+
return nn.Conv3d(*args, **kwargs)
|
231 |
+
raise ValueError(f"unsupported dimensions: {dims}")
|
232 |
+
|
233 |
+
|
234 |
+
def linear(*args, **kwargs):
|
235 |
+
"""
|
236 |
+
Create a linear module.
|
237 |
+
"""
|
238 |
+
return nn.Linear(*args, **kwargs)
|
239 |
+
|
240 |
+
|
241 |
+
def avg_pool_nd(dims, *args, **kwargs):
|
242 |
+
"""
|
243 |
+
Create a 1D, 2D, or 3D average pooling module.
|
244 |
+
"""
|
245 |
+
if dims == 1:
|
246 |
+
return nn.AvgPool1d(*args, **kwargs)
|
247 |
+
elif dims == 2:
|
248 |
+
return nn.AvgPool2d(*args, **kwargs)
|
249 |
+
elif dims == 3:
|
250 |
+
return nn.AvgPool3d(*args, **kwargs)
|
251 |
+
raise ValueError(f"unsupported dimensions: {dims}")
|
252 |
+
|
253 |
+
|
254 |
+
class HybridConditioner(nn.Module):
|
255 |
+
|
256 |
+
def __init__(self, c_concat_config, c_crossattn_config):
|
257 |
+
super().__init__()
|
258 |
+
self.concat_conditioner = instantiate_from_config(c_concat_config)
|
259 |
+
self.crossattn_conditioner = instantiate_from_config(c_crossattn_config)
|
260 |
+
|
261 |
+
def forward(self, c_concat, c_crossattn):
|
262 |
+
c_concat = self.concat_conditioner(c_concat)
|
263 |
+
c_crossattn = self.crossattn_conditioner(c_crossattn)
|
264 |
+
return {'c_concat': [c_concat], 'c_crossattn': [c_crossattn]}
|
265 |
+
|
266 |
+
|
267 |
+
def noise_like(shape, device, repeat=False):
|
268 |
+
repeat_noise = lambda: torch.randn((1, *shape[1:]), device=device).repeat(shape[0], *((1,) * (len(shape) - 1)))
|
269 |
+
noise = lambda: torch.randn(shape, device=device)
|
270 |
+
return repeat_noise() if repeat else noise()
|
ldm/modules/distributions/distributions.py
ADDED
@@ -0,0 +1,92 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import torch
|
2 |
+
import numpy as np
|
3 |
+
|
4 |
+
|
5 |
+
class AbstractDistribution:
|
6 |
+
def sample(self):
|
7 |
+
raise NotImplementedError()
|
8 |
+
|
9 |
+
def mode(self):
|
10 |
+
raise NotImplementedError()
|
11 |
+
|
12 |
+
|
13 |
+
class DiracDistribution(AbstractDistribution):
|
14 |
+
def __init__(self, value):
|
15 |
+
self.value = value
|
16 |
+
|
17 |
+
def sample(self):
|
18 |
+
return self.value
|
19 |
+
|
20 |
+
def mode(self):
|
21 |
+
return self.value
|
22 |
+
|
23 |
+
|
24 |
+
class DiagonalGaussianDistribution(object):
|
25 |
+
def __init__(self, parameters, deterministic=False):
|
26 |
+
self.parameters = parameters
|
27 |
+
self.mean, self.logvar = torch.chunk(parameters, 2, dim=1)
|
28 |
+
self.logvar = torch.clamp(self.logvar, -30.0, 20.0)
|
29 |
+
self.deterministic = deterministic
|
30 |
+
self.std = torch.exp(0.5 * self.logvar)
|
31 |
+
self.var = torch.exp(self.logvar)
|
32 |
+
if self.deterministic:
|
33 |
+
self.var = self.std = torch.zeros_like(self.mean).to(device=self.parameters.device)
|
34 |
+
|
35 |
+
def sample(self):
|
36 |
+
x = self.mean + self.std * torch.randn(self.mean.shape).to(device=self.parameters.device)
|
37 |
+
return x
|
38 |
+
|
39 |
+
def kl(self, other=None):
|
40 |
+
if self.deterministic:
|
41 |
+
return torch.Tensor([0.])
|
42 |
+
else:
|
43 |
+
if other is None:
|
44 |
+
return 0.5 * torch.sum(torch.pow(self.mean, 2)
|
45 |
+
+ self.var - 1.0 - self.logvar,
|
46 |
+
dim=[1, 2, 3])
|
47 |
+
else:
|
48 |
+
return 0.5 * torch.sum(
|
49 |
+
torch.pow(self.mean - other.mean, 2) / other.var
|
50 |
+
+ self.var / other.var - 1.0 - self.logvar + other.logvar,
|
51 |
+
dim=[1, 2, 3])
|
52 |
+
|
53 |
+
def nll(self, sample, dims=[1,2,3]):
|
54 |
+
if self.deterministic:
|
55 |
+
return torch.Tensor([0.])
|
56 |
+
logtwopi = np.log(2.0 * np.pi)
|
57 |
+
return 0.5 * torch.sum(
|
58 |
+
logtwopi + self.logvar + torch.pow(sample - self.mean, 2) / self.var,
|
59 |
+
dim=dims)
|
60 |
+
|
61 |
+
def mode(self):
|
62 |
+
return self.mean
|
63 |
+
|
64 |
+
|
65 |
+
def normal_kl(mean1, logvar1, mean2, logvar2):
|
66 |
+
"""
|
67 |
+
source: https://github.com/openai/guided-diffusion/blob/27c20a8fab9cb472df5d6bdd6c8d11c8f430b924/guided_diffusion/losses.py#L12
|
68 |
+
Compute the KL divergence between two gaussians.
|
69 |
+
Shapes are automatically broadcasted, so batches can be compared to
|
70 |
+
scalars, among other use cases.
|
71 |
+
"""
|
72 |
+
tensor = None
|
73 |
+
for obj in (mean1, logvar1, mean2, logvar2):
|
74 |
+
if isinstance(obj, torch.Tensor):
|
75 |
+
tensor = obj
|
76 |
+
break
|
77 |
+
assert tensor is not None, "at least one argument must be a Tensor"
|
78 |
+
|
79 |
+
# Force variances to be Tensors. Broadcasting helps convert scalars to
|
80 |
+
# Tensors, but it does not work for torch.exp().
|
81 |
+
logvar1, logvar2 = [
|
82 |
+
x if isinstance(x, torch.Tensor) else torch.tensor(x).to(tensor)
|
83 |
+
for x in (logvar1, logvar2)
|
84 |
+
]
|
85 |
+
|
86 |
+
return 0.5 * (
|
87 |
+
-1.0
|
88 |
+
+ logvar2
|
89 |
+
- logvar1
|
90 |
+
+ torch.exp(logvar1 - logvar2)
|
91 |
+
+ ((mean1 - mean2) ** 2) * torch.exp(-logvar2)
|
92 |
+
)
|
ldm/modules/ema.py
ADDED
@@ -0,0 +1,76 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import torch
|
2 |
+
from torch import nn
|
3 |
+
|
4 |
+
|
5 |
+
class LitEma(nn.Module):
|
6 |
+
def __init__(self, model, decay=0.9999, use_num_upates=True):
|
7 |
+
super().__init__()
|
8 |
+
if decay < 0.0 or decay > 1.0:
|
9 |
+
raise ValueError('Decay must be between 0 and 1')
|
10 |
+
|
11 |
+
self.m_name2s_name = {}
|
12 |
+
self.register_buffer('decay', torch.tensor(decay, dtype=torch.float32))
|
13 |
+
self.register_buffer('num_updates', torch.tensor(0,dtype=torch.int) if use_num_upates
|
14 |
+
else torch.tensor(-1,dtype=torch.int))
|
15 |
+
|
16 |
+
for name, p in model.named_parameters():
|
17 |
+
if p.requires_grad:
|
18 |
+
#remove as '.'-character is not allowed in buffers
|
19 |
+
s_name = name.replace('.','')
|
20 |
+
self.m_name2s_name.update({name:s_name})
|
21 |
+
self.register_buffer(s_name,p.clone().detach().data)
|
22 |
+
|
23 |
+
self.collected_params = []
|
24 |
+
|
25 |
+
def forward(self,model):
|
26 |
+
decay = self.decay
|
27 |
+
|
28 |
+
if self.num_updates >= 0:
|
29 |
+
self.num_updates += 1
|
30 |
+
decay = min(self.decay,(1 + self.num_updates) / (10 + self.num_updates))
|
31 |
+
|
32 |
+
one_minus_decay = 1.0 - decay
|
33 |
+
|
34 |
+
with torch.no_grad():
|
35 |
+
m_param = dict(model.named_parameters())
|
36 |
+
shadow_params = dict(self.named_buffers())
|
37 |
+
|
38 |
+
for key in m_param:
|
39 |
+
if m_param[key].requires_grad:
|
40 |
+
sname = self.m_name2s_name[key]
|
41 |
+
shadow_params[sname] = shadow_params[sname].type_as(m_param[key])
|
42 |
+
shadow_params[sname].sub_(one_minus_decay * (shadow_params[sname] - m_param[key]))
|
43 |
+
else:
|
44 |
+
assert not key in self.m_name2s_name
|
45 |
+
|
46 |
+
def copy_to(self, model):
|
47 |
+
m_param = dict(model.named_parameters())
|
48 |
+
shadow_params = dict(self.named_buffers())
|
49 |
+
for key in m_param:
|
50 |
+
if m_param[key].requires_grad:
|
51 |
+
m_param[key].data.copy_(shadow_params[self.m_name2s_name[key]].data)
|
52 |
+
else:
|
53 |
+
assert not key in self.m_name2s_name
|
54 |
+
|
55 |
+
def store(self, parameters):
|
56 |
+
"""
|
57 |
+
Save the current parameters for restoring later.
|
58 |
+
Args:
|
59 |
+
parameters: Iterable of `torch.nn.Parameter`; the parameters to be
|
60 |
+
temporarily stored.
|
61 |
+
"""
|
62 |
+
self.collected_params = [param.clone() for param in parameters]
|
63 |
+
|
64 |
+
def restore(self, parameters):
|
65 |
+
"""
|
66 |
+
Restore the parameters stored with the `store` method.
|
67 |
+
Useful to validate the model with EMA parameters without affecting the
|
68 |
+
original optimization process. Store the parameters before the
|
69 |
+
`copy_to` method. After validation (or model saving), use this to
|
70 |
+
restore the former parameters.
|
71 |
+
Args:
|
72 |
+
parameters: Iterable of `torch.nn.Parameter`; the parameters to be
|
73 |
+
updated with the stored parameters.
|
74 |
+
"""
|
75 |
+
for c_param, param in zip(self.collected_params, parameters):
|
76 |
+
param.data.copy_(c_param.data)
|
ldm/modules/encoders/modules.py
ADDED
@@ -0,0 +1,52 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import torch
|
2 |
+
import torch.nn as nn
|
3 |
+
from ldm.modules.x_transformer import Encoder, TransformerWrapper
|
4 |
+
|
5 |
+
|
6 |
+
class BERTTokenizer(nn.Module):
|
7 |
+
def __init__(self, vq_interface=True, max_length=77):
|
8 |
+
super().__init__()
|
9 |
+
from transformers import BertTokenizerFast
|
10 |
+
self.tokenizer = BertTokenizerFast.from_pretrained("bert-base-uncased")
|
11 |
+
self.vq_interface = vq_interface
|
12 |
+
self.max_length = max_length
|
13 |
+
|
14 |
+
def forward(self, text, return_batch_encoding=False):
|
15 |
+
batch_encoding = self.tokenizer(text, truncation=True, max_length=self.max_length, return_length=True,
|
16 |
+
return_overflowing_tokens=False, padding="max_length", return_tensors="pt")
|
17 |
+
tokens = batch_encoding["input_ids"]
|
18 |
+
if return_batch_encoding:
|
19 |
+
return tokens, batch_encoding
|
20 |
+
return tokens
|
21 |
+
|
22 |
+
@torch.no_grad()
|
23 |
+
def encode(self, text):
|
24 |
+
tokens = self(text)
|
25 |
+
if not self.vq_interface:
|
26 |
+
return tokens
|
27 |
+
return None, None, [None, None, tokens]
|
28 |
+
|
29 |
+
def decode(self, text):
|
30 |
+
return text
|
31 |
+
|
32 |
+
class BERTEmbedder(nn.Module):
|
33 |
+
"""Uses the BERT tokenizer model and adds some transformer encoder layers"""
|
34 |
+
def __init__(self, n_embed, n_layer, vocab_size=30522, max_seq_len=77, use_tokenizer=True, embedding_dropout=0.0):
|
35 |
+
super().__init__()
|
36 |
+
self.use_tknz_fn = use_tokenizer
|
37 |
+
if self.use_tknz_fn:
|
38 |
+
self.tknz_fn = BERTTokenizer(vq_interface=False, max_length=max_seq_len)
|
39 |
+
self.transformer = TransformerWrapper(num_tokens=vocab_size, max_seq_len=max_seq_len,
|
40 |
+
attn_layers=Encoder(dim=n_embed, depth=n_layer),
|
41 |
+
emb_dropout=embedding_dropout)
|
42 |
+
|
43 |
+
def forward(self, cond, text):
|
44 |
+
assert cond is None # Not supported for now (LDM conditioning key == "concat")
|
45 |
+
if self.use_tknz_fn:
|
46 |
+
tokens = self.tknz_fn(text)
|
47 |
+
if next(self.transformer.parameters()).is_cuda:
|
48 |
+
tokens = tokens.cuda()
|
49 |
+
else:
|
50 |
+
tokens = text
|
51 |
+
z = self.transformer(tokens, return_embeddings=True) # Size: [batch_size, max_seq_len, n_embed]
|
52 |
+
return z
|
ldm/modules/losses/contperceptual.py
ADDED
@@ -0,0 +1,111 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import torch
|
2 |
+
import torch.nn as nn
|
3 |
+
|
4 |
+
from taming.modules.losses.vqperceptual import * # TODO: taming dependency yes/no?
|
5 |
+
|
6 |
+
|
7 |
+
class LPIPSWithDiscriminator(nn.Module):
|
8 |
+
def __init__(self, disc_start, logvar_init=0.0, kl_weight=1.0, pixelloss_weight=1.0,
|
9 |
+
disc_num_layers=3, disc_in_channels=3, disc_factor=1.0, disc_weight=1.0,
|
10 |
+
perceptual_weight=1.0, use_actnorm=False, disc_conditional=False,
|
11 |
+
disc_loss="hinge"):
|
12 |
+
|
13 |
+
super().__init__()
|
14 |
+
assert disc_loss in ["hinge", "vanilla"]
|
15 |
+
self.kl_weight = kl_weight
|
16 |
+
self.pixel_weight = pixelloss_weight
|
17 |
+
self.perceptual_loss = LPIPS().eval()
|
18 |
+
self.perceptual_weight = perceptual_weight
|
19 |
+
# output log variance
|
20 |
+
self.logvar = nn.Parameter(torch.ones(size=()) * logvar_init)
|
21 |
+
|
22 |
+
self.discriminator = NLayerDiscriminator(input_nc=disc_in_channels,
|
23 |
+
n_layers=disc_num_layers,
|
24 |
+
use_actnorm=use_actnorm
|
25 |
+
).apply(weights_init)
|
26 |
+
self.discriminator_iter_start = disc_start
|
27 |
+
self.disc_loss = hinge_d_loss if disc_loss == "hinge" else vanilla_d_loss
|
28 |
+
self.disc_factor = disc_factor
|
29 |
+
self.discriminator_weight = disc_weight
|
30 |
+
self.disc_conditional = disc_conditional
|
31 |
+
|
32 |
+
def calculate_adaptive_weight(self, nll_loss, g_loss, last_layer=None):
|
33 |
+
if last_layer is not None:
|
34 |
+
nll_grads = torch.autograd.grad(nll_loss, last_layer, retain_graph=True)[0]
|
35 |
+
g_grads = torch.autograd.grad(g_loss, last_layer, retain_graph=True)[0]
|
36 |
+
else:
|
37 |
+
nll_grads = torch.autograd.grad(nll_loss, self.last_layer[0], retain_graph=True)[0]
|
38 |
+
g_grads = torch.autograd.grad(g_loss, self.last_layer[0], retain_graph=True)[0]
|
39 |
+
|
40 |
+
d_weight = torch.norm(nll_grads) / (torch.norm(g_grads) + 1e-4)
|
41 |
+
d_weight = torch.clamp(d_weight, 0.0, 1e4).detach()
|
42 |
+
d_weight = d_weight * self.discriminator_weight
|
43 |
+
return d_weight
|
44 |
+
|
45 |
+
def forward(self, inputs, reconstructions, posteriors, optimizer_idx,
|
46 |
+
global_step, last_layer=None, cond=None, split="train",
|
47 |
+
weights=None):
|
48 |
+
rec_loss = torch.abs(inputs.contiguous() - reconstructions.contiguous())
|
49 |
+
if self.perceptual_weight > 0:
|
50 |
+
p_loss = self.perceptual_loss(inputs.contiguous(), reconstructions.contiguous())
|
51 |
+
rec_loss = rec_loss + self.perceptual_weight * p_loss
|
52 |
+
|
53 |
+
nll_loss = rec_loss / torch.exp(self.logvar) + self.logvar
|
54 |
+
weighted_nll_loss = nll_loss
|
55 |
+
if weights is not None:
|
56 |
+
weighted_nll_loss = weights*nll_loss
|
57 |
+
weighted_nll_loss = torch.sum(weighted_nll_loss) / weighted_nll_loss.shape[0]
|
58 |
+
nll_loss = torch.sum(nll_loss) / nll_loss.shape[0]
|
59 |
+
kl_loss = posteriors.kl()
|
60 |
+
kl_loss = torch.sum(kl_loss) / kl_loss.shape[0]
|
61 |
+
|
62 |
+
# now the GAN part
|
63 |
+
if optimizer_idx == 0:
|
64 |
+
# generator update
|
65 |
+
if cond is None:
|
66 |
+
assert not self.disc_conditional
|
67 |
+
logits_fake = self.discriminator(reconstructions.contiguous())
|
68 |
+
else:
|
69 |
+
assert self.disc_conditional
|
70 |
+
logits_fake = self.discriminator(torch.cat((reconstructions.contiguous(), cond), dim=1))
|
71 |
+
g_loss = -torch.mean(logits_fake)
|
72 |
+
|
73 |
+
if self.disc_factor > 0.0:
|
74 |
+
try:
|
75 |
+
d_weight = self.calculate_adaptive_weight(nll_loss, g_loss, last_layer=last_layer)
|
76 |
+
except RuntimeError:
|
77 |
+
assert not self.training
|
78 |
+
d_weight = torch.tensor(0.0)
|
79 |
+
else:
|
80 |
+
d_weight = torch.tensor(0.0)
|
81 |
+
|
82 |
+
disc_factor = adopt_weight(self.disc_factor, global_step, threshold=self.discriminator_iter_start)
|
83 |
+
loss = weighted_nll_loss + self.kl_weight * kl_loss + d_weight * disc_factor * g_loss
|
84 |
+
|
85 |
+
log = {"{}/total_loss".format(split): loss.clone().detach().mean(), "{}/logvar".format(split): self.logvar.detach(),
|
86 |
+
"{}/kl_loss".format(split): kl_loss.detach().mean(), "{}/nll_loss".format(split): nll_loss.detach().mean(),
|
87 |
+
"{}/rec_loss".format(split): rec_loss.detach().mean(),
|
88 |
+
"{}/d_weight".format(split): d_weight.detach(),
|
89 |
+
"{}/disc_factor".format(split): torch.tensor(disc_factor),
|
90 |
+
"{}/g_loss".format(split): g_loss.detach().mean(),
|
91 |
+
}
|
92 |
+
return loss, log
|
93 |
+
|
94 |
+
if optimizer_idx == 1:
|
95 |
+
# second pass for discriminator update
|
96 |
+
if cond is None:
|
97 |
+
logits_real = self.discriminator(inputs.contiguous().detach())
|
98 |
+
logits_fake = self.discriminator(reconstructions.contiguous().detach())
|
99 |
+
else:
|
100 |
+
logits_real = self.discriminator(torch.cat((inputs.contiguous().detach(), cond), dim=1))
|
101 |
+
logits_fake = self.discriminator(torch.cat((reconstructions.contiguous().detach(), cond), dim=1))
|
102 |
+
|
103 |
+
disc_factor = adopt_weight(self.disc_factor, global_step, threshold=self.discriminator_iter_start)
|
104 |
+
d_loss = disc_factor * self.disc_loss(logits_real, logits_fake)
|
105 |
+
|
106 |
+
log = {"{}/disc_loss".format(split): d_loss.clone().detach().mean(),
|
107 |
+
"{}/logits_real".format(split): logits_real.detach().mean(),
|
108 |
+
"{}/logits_fake".format(split): logits_fake.detach().mean()
|
109 |
+
}
|
110 |
+
return d_loss, log
|
111 |
+
|
ldm/modules/losses/vqperceptual.py
ADDED
@@ -0,0 +1,167 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import torch
|
2 |
+
from torch import nn
|
3 |
+
import torch.nn.functional as F
|
4 |
+
from einops import repeat
|
5 |
+
|
6 |
+
from taming.modules.discriminator.model import NLayerDiscriminator, weights_init
|
7 |
+
from taming.modules.losses.lpips import LPIPS
|
8 |
+
from taming.modules.losses.vqperceptual import hinge_d_loss, vanilla_d_loss
|
9 |
+
|
10 |
+
|
11 |
+
def hinge_d_loss_with_exemplar_weights(logits_real, logits_fake, weights):
|
12 |
+
assert weights.shape[0] == logits_real.shape[0] == logits_fake.shape[0]
|
13 |
+
loss_real = torch.mean(F.relu(1. - logits_real), dim=[1,2,3])
|
14 |
+
loss_fake = torch.mean(F.relu(1. + logits_fake), dim=[1,2,3])
|
15 |
+
loss_real = (weights * loss_real).sum() / weights.sum()
|
16 |
+
loss_fake = (weights * loss_fake).sum() / weights.sum()
|
17 |
+
d_loss = 0.5 * (loss_real + loss_fake)
|
18 |
+
return d_loss
|
19 |
+
|
20 |
+
def adopt_weight(weight, global_step, threshold=0, value=0.):
|
21 |
+
if global_step < threshold:
|
22 |
+
weight = value
|
23 |
+
return weight
|
24 |
+
|
25 |
+
|
26 |
+
def measure_perplexity(predicted_indices, n_embed):
|
27 |
+
# src: https://github.com/karpathy/deep-vector-quantization/blob/main/model.py
|
28 |
+
# eval cluster perplexity. when perplexity == num_embeddings then all clusters are used exactly equally
|
29 |
+
encodings = F.one_hot(predicted_indices, n_embed).float().reshape(-1, n_embed)
|
30 |
+
avg_probs = encodings.mean(0)
|
31 |
+
perplexity = (-(avg_probs * torch.log(avg_probs + 1e-10)).sum()).exp()
|
32 |
+
cluster_use = torch.sum(avg_probs > 0)
|
33 |
+
return perplexity, cluster_use
|
34 |
+
|
35 |
+
def l1(x, y):
|
36 |
+
return torch.abs(x-y)
|
37 |
+
|
38 |
+
|
39 |
+
def l2(x, y):
|
40 |
+
return torch.pow((x-y), 2)
|
41 |
+
|
42 |
+
|
43 |
+
class VQLPIPSWithDiscriminator(nn.Module):
|
44 |
+
def __init__(self, disc_start, codebook_weight=1.0, pixelloss_weight=1.0,
|
45 |
+
disc_num_layers=3, disc_in_channels=3, disc_factor=1.0, disc_weight=1.0,
|
46 |
+
perceptual_weight=1.0, use_actnorm=False, disc_conditional=False,
|
47 |
+
disc_ndf=64, disc_loss="hinge", n_classes=None, perceptual_loss="lpips",
|
48 |
+
pixel_loss="l1"):
|
49 |
+
super().__init__()
|
50 |
+
assert disc_loss in ["hinge", "vanilla"]
|
51 |
+
assert perceptual_loss in ["lpips", "clips", "dists"]
|
52 |
+
assert pixel_loss in ["l1", "l2"]
|
53 |
+
self.codebook_weight = codebook_weight
|
54 |
+
self.pixel_weight = pixelloss_weight
|
55 |
+
if perceptual_loss == "lpips":
|
56 |
+
print(f"{self.__class__.__name__}: Running with LPIPS.")
|
57 |
+
self.perceptual_loss = LPIPS().eval()
|
58 |
+
else:
|
59 |
+
raise ValueError(f"Unknown perceptual loss: >> {perceptual_loss} <<")
|
60 |
+
self.perceptual_weight = perceptual_weight
|
61 |
+
|
62 |
+
if pixel_loss == "l1":
|
63 |
+
self.pixel_loss = l1
|
64 |
+
else:
|
65 |
+
self.pixel_loss = l2
|
66 |
+
|
67 |
+
self.discriminator = NLayerDiscriminator(input_nc=disc_in_channels,
|
68 |
+
n_layers=disc_num_layers,
|
69 |
+
use_actnorm=use_actnorm,
|
70 |
+
ndf=disc_ndf
|
71 |
+
).apply(weights_init)
|
72 |
+
self.discriminator_iter_start = disc_start
|
73 |
+
if disc_loss == "hinge":
|
74 |
+
self.disc_loss = hinge_d_loss
|
75 |
+
elif disc_loss == "vanilla":
|
76 |
+
self.disc_loss = vanilla_d_loss
|
77 |
+
else:
|
78 |
+
raise ValueError(f"Unknown GAN loss '{disc_loss}'.")
|
79 |
+
print(f"VQLPIPSWithDiscriminator running with {disc_loss} loss.")
|
80 |
+
self.disc_factor = disc_factor
|
81 |
+
self.discriminator_weight = disc_weight
|
82 |
+
self.disc_conditional = disc_conditional
|
83 |
+
self.n_classes = n_classes
|
84 |
+
|
85 |
+
def calculate_adaptive_weight(self, nll_loss, g_loss, last_layer=None):
|
86 |
+
if last_layer is not None:
|
87 |
+
nll_grads = torch.autograd.grad(nll_loss, last_layer, retain_graph=True)[0]
|
88 |
+
g_grads = torch.autograd.grad(g_loss, last_layer, retain_graph=True)[0]
|
89 |
+
else:
|
90 |
+
nll_grads = torch.autograd.grad(nll_loss, self.last_layer[0], retain_graph=True)[0]
|
91 |
+
g_grads = torch.autograd.grad(g_loss, self.last_layer[0], retain_graph=True)[0]
|
92 |
+
|
93 |
+
d_weight = torch.norm(nll_grads) / (torch.norm(g_grads) + 1e-4)
|
94 |
+
d_weight = torch.clamp(d_weight, 0.0, 1e4).detach()
|
95 |
+
d_weight = d_weight * self.discriminator_weight
|
96 |
+
return d_weight
|
97 |
+
|
98 |
+
def forward(self, codebook_loss, inputs, reconstructions, optimizer_idx,
|
99 |
+
global_step, last_layer=None, cond=None, split="train", predicted_indices=None):
|
100 |
+
if codebook_loss is None:
|
101 |
+
codebook_loss = torch.tensor([0.]).to(inputs.device)
|
102 |
+
#rec_loss = torch.abs(inputs.contiguous() - reconstructions.contiguous())
|
103 |
+
rec_loss = self.pixel_loss(inputs.contiguous(), reconstructions.contiguous())
|
104 |
+
if self.perceptual_weight > 0:
|
105 |
+
p_loss = self.perceptual_loss(inputs.contiguous(), reconstructions.contiguous())
|
106 |
+
rec_loss = rec_loss + self.perceptual_weight * p_loss
|
107 |
+
else:
|
108 |
+
p_loss = torch.tensor([0.0])
|
109 |
+
|
110 |
+
nll_loss = rec_loss
|
111 |
+
#nll_loss = torch.sum(nll_loss) / nll_loss.shape[0]
|
112 |
+
nll_loss = torch.mean(nll_loss)
|
113 |
+
|
114 |
+
# now the GAN part
|
115 |
+
if optimizer_idx == 0:
|
116 |
+
# generator update
|
117 |
+
if cond is None:
|
118 |
+
assert not self.disc_conditional
|
119 |
+
logits_fake = self.discriminator(reconstructions.contiguous())
|
120 |
+
else:
|
121 |
+
assert self.disc_conditional
|
122 |
+
logits_fake = self.discriminator(torch.cat((reconstructions.contiguous(), cond), dim=1))
|
123 |
+
g_loss = -torch.mean(logits_fake)
|
124 |
+
|
125 |
+
try:
|
126 |
+
d_weight = self.calculate_adaptive_weight(nll_loss, g_loss, last_layer=last_layer)
|
127 |
+
except RuntimeError:
|
128 |
+
assert not self.training
|
129 |
+
d_weight = torch.tensor(0.0)
|
130 |
+
|
131 |
+
disc_factor = adopt_weight(self.disc_factor, global_step, threshold=self.discriminator_iter_start)
|
132 |
+
loss = nll_loss + d_weight * disc_factor * g_loss + self.codebook_weight * codebook_loss.mean()
|
133 |
+
|
134 |
+
log = {"{}/total_loss".format(split): loss.clone().detach().mean(),
|
135 |
+
"{}/quant_loss".format(split): codebook_loss.detach().mean(),
|
136 |
+
"{}/nll_loss".format(split): nll_loss.detach().mean(),
|
137 |
+
"{}/rec_loss".format(split): rec_loss.detach().mean(),
|
138 |
+
"{}/p_loss".format(split): p_loss.detach().mean(),
|
139 |
+
"{}/d_weight".format(split): d_weight.detach(),
|
140 |
+
"{}/disc_factor".format(split): torch.tensor(disc_factor),
|
141 |
+
"{}/g_loss".format(split): g_loss.detach().mean(),
|
142 |
+
}
|
143 |
+
# if predicted_indices is not None:
|
144 |
+
# assert self.n_classes is not None
|
145 |
+
# with torch.no_grad():
|
146 |
+
# perplexity, cluster_usage = measure_perplexity(predicted_indices, self.n_classes)
|
147 |
+
# log[f"{split}/perplexity"] = perplexity
|
148 |
+
# log[f"{split}/cluster_usage"] = cluster_usage
|
149 |
+
return loss, log
|
150 |
+
|
151 |
+
if optimizer_idx == 1:
|
152 |
+
# second pass for discriminator update
|
153 |
+
if cond is None:
|
154 |
+
logits_real = self.discriminator(inputs.contiguous().detach())
|
155 |
+
logits_fake = self.discriminator(reconstructions.contiguous().detach())
|
156 |
+
else:
|
157 |
+
logits_real = self.discriminator(torch.cat((inputs.contiguous().detach(), cond), dim=1))
|
158 |
+
logits_fake = self.discriminator(torch.cat((reconstructions.contiguous().detach(), cond), dim=1))
|
159 |
+
|
160 |
+
disc_factor = adopt_weight(self.disc_factor, global_step, threshold=self.discriminator_iter_start)
|
161 |
+
d_loss = disc_factor * self.disc_loss(logits_real, logits_fake)
|
162 |
+
|
163 |
+
log = {"{}/disc_loss".format(split): d_loss.clone().detach().mean(),
|
164 |
+
"{}/logits_real".format(split): logits_real.detach().mean(),
|
165 |
+
"{}/logits_fake".format(split): logits_fake.detach().mean()
|
166 |
+
}
|
167 |
+
return d_loss, log
|
ldm/modules/x_transformer.py
ADDED
@@ -0,0 +1,641 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
"""shout-out to https://github.com/lucidrains/x-transformers/tree/main/x_transformers"""
|
2 |
+
import torch
|
3 |
+
from torch import nn, einsum
|
4 |
+
import torch.nn.functional as F
|
5 |
+
from functools import partial
|
6 |
+
from inspect import isfunction
|
7 |
+
from collections import namedtuple
|
8 |
+
from einops import rearrange, repeat, reduce
|
9 |
+
|
10 |
+
# constants
|
11 |
+
|
12 |
+
DEFAULT_DIM_HEAD = 64
|
13 |
+
|
14 |
+
Intermediates = namedtuple('Intermediates', [
|
15 |
+
'pre_softmax_attn',
|
16 |
+
'post_softmax_attn'
|
17 |
+
])
|
18 |
+
|
19 |
+
LayerIntermediates = namedtuple('Intermediates', [
|
20 |
+
'hiddens',
|
21 |
+
'attn_intermediates'
|
22 |
+
])
|
23 |
+
|
24 |
+
|
25 |
+
class AbsolutePositionalEmbedding(nn.Module):
|
26 |
+
def __init__(self, dim, max_seq_len):
|
27 |
+
super().__init__()
|
28 |
+
self.emb = nn.Embedding(max_seq_len, dim)
|
29 |
+
self.init_()
|
30 |
+
|
31 |
+
def init_(self):
|
32 |
+
nn.init.normal_(self.emb.weight, std=0.02)
|
33 |
+
|
34 |
+
def forward(self, x):
|
35 |
+
n = torch.arange(x.shape[1], device=x.device)
|
36 |
+
return self.emb(n)[None, :, :]
|
37 |
+
|
38 |
+
|
39 |
+
class FixedPositionalEmbedding(nn.Module):
|
40 |
+
def __init__(self, dim):
|
41 |
+
super().__init__()
|
42 |
+
inv_freq = 1. / (10000 ** (torch.arange(0, dim, 2).float() / dim))
|
43 |
+
self.register_buffer('inv_freq', inv_freq)
|
44 |
+
|
45 |
+
def forward(self, x, seq_dim=1, offset=0):
|
46 |
+
t = torch.arange(x.shape[seq_dim], device=x.device).type_as(self.inv_freq) + offset
|
47 |
+
sinusoid_inp = torch.einsum('i , j -> i j', t, self.inv_freq)
|
48 |
+
emb = torch.cat((sinusoid_inp.sin(), sinusoid_inp.cos()), dim=-1)
|
49 |
+
return emb[None, :, :]
|
50 |
+
|
51 |
+
|
52 |
+
# helpers
|
53 |
+
|
54 |
+
def exists(val):
|
55 |
+
return val is not None
|
56 |
+
|
57 |
+
|
58 |
+
def default(val, d):
|
59 |
+
if exists(val):
|
60 |
+
return val
|
61 |
+
return d() if isfunction(d) else d
|
62 |
+
|
63 |
+
|
64 |
+
def always(val):
|
65 |
+
def inner(*args, **kwargs):
|
66 |
+
return val
|
67 |
+
return inner
|
68 |
+
|
69 |
+
|
70 |
+
def not_equals(val):
|
71 |
+
def inner(x):
|
72 |
+
return x != val
|
73 |
+
return inner
|
74 |
+
|
75 |
+
|
76 |
+
def equals(val):
|
77 |
+
def inner(x):
|
78 |
+
return x == val
|
79 |
+
return inner
|
80 |
+
|
81 |
+
|
82 |
+
def max_neg_value(tensor):
|
83 |
+
return -torch.finfo(tensor.dtype).max
|
84 |
+
|
85 |
+
|
86 |
+
# keyword argument helpers
|
87 |
+
|
88 |
+
def pick_and_pop(keys, d):
|
89 |
+
values = list(map(lambda key: d.pop(key), keys))
|
90 |
+
return dict(zip(keys, values))
|
91 |
+
|
92 |
+
|
93 |
+
def group_dict_by_key(cond, d):
|
94 |
+
return_val = [dict(), dict()]
|
95 |
+
for key in d.keys():
|
96 |
+
match = bool(cond(key))
|
97 |
+
ind = int(not match)
|
98 |
+
return_val[ind][key] = d[key]
|
99 |
+
return (*return_val,)
|
100 |
+
|
101 |
+
|
102 |
+
def string_begins_with(prefix, str):
|
103 |
+
return str.startswith(prefix)
|
104 |
+
|
105 |
+
|
106 |
+
def group_by_key_prefix(prefix, d):
|
107 |
+
return group_dict_by_key(partial(string_begins_with, prefix), d)
|
108 |
+
|
109 |
+
|
110 |
+
def groupby_prefix_and_trim(prefix, d):
|
111 |
+
kwargs_with_prefix, kwargs = group_dict_by_key(partial(string_begins_with, prefix), d)
|
112 |
+
kwargs_without_prefix = dict(map(lambda x: (x[0][len(prefix):], x[1]), tuple(kwargs_with_prefix.items())))
|
113 |
+
return kwargs_without_prefix, kwargs
|
114 |
+
|
115 |
+
|
116 |
+
# classes
|
117 |
+
class Scale(nn.Module):
|
118 |
+
def __init__(self, value, fn):
|
119 |
+
super().__init__()
|
120 |
+
self.value = value
|
121 |
+
self.fn = fn
|
122 |
+
|
123 |
+
def forward(self, x, **kwargs):
|
124 |
+
x, *rest = self.fn(x, **kwargs)
|
125 |
+
return (x * self.value, *rest)
|
126 |
+
|
127 |
+
|
128 |
+
class Rezero(nn.Module):
|
129 |
+
def __init__(self, fn):
|
130 |
+
super().__init__()
|
131 |
+
self.fn = fn
|
132 |
+
self.g = nn.Parameter(torch.zeros(1))
|
133 |
+
|
134 |
+
def forward(self, x, **kwargs):
|
135 |
+
x, *rest = self.fn(x, **kwargs)
|
136 |
+
return (x * self.g, *rest)
|
137 |
+
|
138 |
+
|
139 |
+
class ScaleNorm(nn.Module):
|
140 |
+
def __init__(self, dim, eps=1e-5):
|
141 |
+
super().__init__()
|
142 |
+
self.scale = dim ** -0.5
|
143 |
+
self.eps = eps
|
144 |
+
self.g = nn.Parameter(torch.ones(1))
|
145 |
+
|
146 |
+
def forward(self, x):
|
147 |
+
norm = torch.norm(x, dim=-1, keepdim=True) * self.scale
|
148 |
+
return x / norm.clamp(min=self.eps) * self.g
|
149 |
+
|
150 |
+
|
151 |
+
class RMSNorm(nn.Module):
|
152 |
+
def __init__(self, dim, eps=1e-8):
|
153 |
+
super().__init__()
|
154 |
+
self.scale = dim ** -0.5
|
155 |
+
self.eps = eps
|
156 |
+
self.g = nn.Parameter(torch.ones(dim))
|
157 |
+
|
158 |
+
def forward(self, x):
|
159 |
+
norm = torch.norm(x, dim=-1, keepdim=True) * self.scale
|
160 |
+
return x / norm.clamp(min=self.eps) * self.g
|
161 |
+
|
162 |
+
|
163 |
+
class Residual(nn.Module):
|
164 |
+
def forward(self, x, residual):
|
165 |
+
return x + residual
|
166 |
+
|
167 |
+
|
168 |
+
class GRUGating(nn.Module):
|
169 |
+
def __init__(self, dim):
|
170 |
+
super().__init__()
|
171 |
+
self.gru = nn.GRUCell(dim, dim)
|
172 |
+
|
173 |
+
def forward(self, x, residual):
|
174 |
+
gated_output = self.gru(
|
175 |
+
rearrange(x, 'b n d -> (b n) d'),
|
176 |
+
rearrange(residual, 'b n d -> (b n) d')
|
177 |
+
)
|
178 |
+
|
179 |
+
return gated_output.reshape_as(x)
|
180 |
+
|
181 |
+
|
182 |
+
# feedforward
|
183 |
+
|
184 |
+
class GEGLU(nn.Module):
|
185 |
+
def __init__(self, dim_in, dim_out):
|
186 |
+
super().__init__()
|
187 |
+
self.proj = nn.Linear(dim_in, dim_out * 2)
|
188 |
+
|
189 |
+
def forward(self, x):
|
190 |
+
x, gate = self.proj(x).chunk(2, dim=-1)
|
191 |
+
return x * F.gelu(gate)
|
192 |
+
|
193 |
+
|
194 |
+
class FeedForward(nn.Module):
|
195 |
+
def __init__(self, dim, dim_out=None, mult=4, glu=False, dropout=0.):
|
196 |
+
super().__init__()
|
197 |
+
inner_dim = int(dim * mult)
|
198 |
+
dim_out = default(dim_out, dim)
|
199 |
+
project_in = nn.Sequential(
|
200 |
+
nn.Linear(dim, inner_dim),
|
201 |
+
nn.GELU()
|
202 |
+
) if not glu else GEGLU(dim, inner_dim)
|
203 |
+
|
204 |
+
self.net = nn.Sequential(
|
205 |
+
project_in,
|
206 |
+
nn.Dropout(dropout),
|
207 |
+
nn.Linear(inner_dim, dim_out)
|
208 |
+
)
|
209 |
+
|
210 |
+
def forward(self, x):
|
211 |
+
return self.net(x)
|
212 |
+
|
213 |
+
|
214 |
+
# attention.
|
215 |
+
class Attention(nn.Module):
|
216 |
+
def __init__(
|
217 |
+
self,
|
218 |
+
dim,
|
219 |
+
dim_head=DEFAULT_DIM_HEAD,
|
220 |
+
heads=8,
|
221 |
+
causal=False,
|
222 |
+
mask=None,
|
223 |
+
talking_heads=False,
|
224 |
+
sparse_topk=None,
|
225 |
+
use_entmax15=False,
|
226 |
+
num_mem_kv=0,
|
227 |
+
dropout=0.,
|
228 |
+
on_attn=False
|
229 |
+
):
|
230 |
+
super().__init__()
|
231 |
+
if use_entmax15:
|
232 |
+
raise NotImplementedError("Check out entmax activation instead of softmax activation!")
|
233 |
+
self.scale = dim_head ** -0.5
|
234 |
+
self.heads = heads
|
235 |
+
self.causal = causal
|
236 |
+
self.mask = mask
|
237 |
+
|
238 |
+
inner_dim = dim_head * heads
|
239 |
+
|
240 |
+
self.to_q = nn.Linear(dim, inner_dim, bias=False)
|
241 |
+
self.to_k = nn.Linear(dim, inner_dim, bias=False)
|
242 |
+
self.to_v = nn.Linear(dim, inner_dim, bias=False)
|
243 |
+
self.dropout = nn.Dropout(dropout)
|
244 |
+
|
245 |
+
# talking heads
|
246 |
+
self.talking_heads = talking_heads
|
247 |
+
if talking_heads:
|
248 |
+
self.pre_softmax_proj = nn.Parameter(torch.randn(heads, heads))
|
249 |
+
self.post_softmax_proj = nn.Parameter(torch.randn(heads, heads))
|
250 |
+
|
251 |
+
# explicit topk sparse attention
|
252 |
+
self.sparse_topk = sparse_topk
|
253 |
+
|
254 |
+
# entmax
|
255 |
+
#self.attn_fn = entmax15 if use_entmax15 else F.softmax
|
256 |
+
self.attn_fn = F.softmax
|
257 |
+
|
258 |
+
# add memory key / values
|
259 |
+
self.num_mem_kv = num_mem_kv
|
260 |
+
if num_mem_kv > 0:
|
261 |
+
self.mem_k = nn.Parameter(torch.randn(heads, num_mem_kv, dim_head))
|
262 |
+
self.mem_v = nn.Parameter(torch.randn(heads, num_mem_kv, dim_head))
|
263 |
+
|
264 |
+
# attention on attention
|
265 |
+
self.attn_on_attn = on_attn
|
266 |
+
self.to_out = nn.Sequential(nn.Linear(inner_dim, dim * 2), nn.GLU()) if on_attn else nn.Linear(inner_dim, dim)
|
267 |
+
|
268 |
+
def forward(
|
269 |
+
self,
|
270 |
+
x,
|
271 |
+
context=None,
|
272 |
+
mask=None,
|
273 |
+
context_mask=None,
|
274 |
+
rel_pos=None,
|
275 |
+
sinusoidal_emb=None,
|
276 |
+
prev_attn=None,
|
277 |
+
mem=None
|
278 |
+
):
|
279 |
+
b, n, _, h, talking_heads, device = *x.shape, self.heads, self.talking_heads, x.device
|
280 |
+
kv_input = default(context, x)
|
281 |
+
|
282 |
+
q_input = x
|
283 |
+
k_input = kv_input
|
284 |
+
v_input = kv_input
|
285 |
+
|
286 |
+
if exists(mem):
|
287 |
+
k_input = torch.cat((mem, k_input), dim=-2)
|
288 |
+
v_input = torch.cat((mem, v_input), dim=-2)
|
289 |
+
|
290 |
+
if exists(sinusoidal_emb):
|
291 |
+
# in shortformer, the query would start at a position offset depending on the past cached memory
|
292 |
+
offset = k_input.shape[-2] - q_input.shape[-2]
|
293 |
+
q_input = q_input + sinusoidal_emb(q_input, offset=offset)
|
294 |
+
k_input = k_input + sinusoidal_emb(k_input)
|
295 |
+
|
296 |
+
q = self.to_q(q_input)
|
297 |
+
k = self.to_k(k_input)
|
298 |
+
v = self.to_v(v_input)
|
299 |
+
|
300 |
+
q, k, v = map(lambda t: rearrange(t, 'b n (h d) -> b h n d', h=h), (q, k, v))
|
301 |
+
|
302 |
+
input_mask = None
|
303 |
+
if any(map(exists, (mask, context_mask))):
|
304 |
+
q_mask = default(mask, lambda: torch.ones((b, n), device=device).bool())
|
305 |
+
k_mask = q_mask if not exists(context) else context_mask
|
306 |
+
k_mask = default(k_mask, lambda: torch.ones((b, k.shape[-2]), device=device).bool())
|
307 |
+
q_mask = rearrange(q_mask, 'b i -> b () i ()')
|
308 |
+
k_mask = rearrange(k_mask, 'b j -> b () () j')
|
309 |
+
input_mask = q_mask * k_mask
|
310 |
+
|
311 |
+
if self.num_mem_kv > 0:
|
312 |
+
mem_k, mem_v = map(lambda t: repeat(t, 'h n d -> b h n d', b=b), (self.mem_k, self.mem_v))
|
313 |
+
k = torch.cat((mem_k, k), dim=-2)
|
314 |
+
v = torch.cat((mem_v, v), dim=-2)
|
315 |
+
if exists(input_mask):
|
316 |
+
input_mask = F.pad(input_mask, (self.num_mem_kv, 0), value=True)
|
317 |
+
|
318 |
+
dots = einsum('b h i d, b h j d -> b h i j', q, k) * self.scale
|
319 |
+
mask_value = max_neg_value(dots)
|
320 |
+
|
321 |
+
if exists(prev_attn):
|
322 |
+
dots = dots + prev_attn
|
323 |
+
|
324 |
+
pre_softmax_attn = dots
|
325 |
+
|
326 |
+
if talking_heads:
|
327 |
+
dots = einsum('b h i j, h k -> b k i j', dots, self.pre_softmax_proj).contiguous()
|
328 |
+
|
329 |
+
if exists(rel_pos):
|
330 |
+
dots = rel_pos(dots)
|
331 |
+
|
332 |
+
if exists(input_mask):
|
333 |
+
dots.masked_fill_(~input_mask, mask_value)
|
334 |
+
del input_mask
|
335 |
+
|
336 |
+
if self.causal:
|
337 |
+
i, j = dots.shape[-2:]
|
338 |
+
r = torch.arange(i, device=device)
|
339 |
+
mask = rearrange(r, 'i -> () () i ()') < rearrange(r, 'j -> () () () j')
|
340 |
+
mask = F.pad(mask, (j - i, 0), value=False)
|
341 |
+
dots.masked_fill_(mask, mask_value)
|
342 |
+
del mask
|
343 |
+
|
344 |
+
if exists(self.sparse_topk) and self.sparse_topk < dots.shape[-1]:
|
345 |
+
top, _ = dots.topk(self.sparse_topk, dim=-1)
|
346 |
+
vk = top[..., -1].unsqueeze(-1).expand_as(dots)
|
347 |
+
mask = dots < vk
|
348 |
+
dots.masked_fill_(mask, mask_value)
|
349 |
+
del mask
|
350 |
+
|
351 |
+
attn = self.attn_fn(dots, dim=-1)
|
352 |
+
post_softmax_attn = attn
|
353 |
+
|
354 |
+
attn = self.dropout(attn)
|
355 |
+
|
356 |
+
if talking_heads:
|
357 |
+
attn = einsum('b h i j, h k -> b k i j', attn, self.post_softmax_proj).contiguous()
|
358 |
+
|
359 |
+
out = einsum('b h i j, b h j d -> b h i d', attn, v)
|
360 |
+
out = rearrange(out, 'b h n d -> b n (h d)')
|
361 |
+
|
362 |
+
intermediates = Intermediates(
|
363 |
+
pre_softmax_attn=pre_softmax_attn,
|
364 |
+
post_softmax_attn=post_softmax_attn
|
365 |
+
)
|
366 |
+
|
367 |
+
return self.to_out(out), intermediates
|
368 |
+
|
369 |
+
|
370 |
+
class AttentionLayers(nn.Module):
|
371 |
+
def __init__(
|
372 |
+
self,
|
373 |
+
dim,
|
374 |
+
depth,
|
375 |
+
heads=8,
|
376 |
+
causal=False,
|
377 |
+
cross_attend=False,
|
378 |
+
only_cross=False,
|
379 |
+
use_scalenorm=False,
|
380 |
+
use_rmsnorm=False,
|
381 |
+
use_rezero=False,
|
382 |
+
rel_pos_num_buckets=32,
|
383 |
+
rel_pos_max_distance=128,
|
384 |
+
position_infused_attn=False,
|
385 |
+
custom_layers=None,
|
386 |
+
sandwich_coef=None,
|
387 |
+
par_ratio=None,
|
388 |
+
residual_attn=False,
|
389 |
+
cross_residual_attn=False,
|
390 |
+
macaron=False,
|
391 |
+
pre_norm=True,
|
392 |
+
gate_residual=False,
|
393 |
+
**kwargs
|
394 |
+
):
|
395 |
+
super().__init__()
|
396 |
+
ff_kwargs, kwargs = groupby_prefix_and_trim('ff_', kwargs)
|
397 |
+
attn_kwargs, _ = groupby_prefix_and_trim('attn_', kwargs)
|
398 |
+
|
399 |
+
dim_head = attn_kwargs.get('dim_head', DEFAULT_DIM_HEAD)
|
400 |
+
|
401 |
+
self.dim = dim
|
402 |
+
self.depth = depth
|
403 |
+
self.layers = nn.ModuleList([])
|
404 |
+
|
405 |
+
self.has_pos_emb = position_infused_attn
|
406 |
+
self.pia_pos_emb = FixedPositionalEmbedding(dim) if position_infused_attn else None
|
407 |
+
self.rotary_pos_emb = always(None)
|
408 |
+
|
409 |
+
assert rel_pos_num_buckets <= rel_pos_max_distance, 'number of relative position buckets must be less than the relative position max distance'
|
410 |
+
self.rel_pos = None
|
411 |
+
|
412 |
+
self.pre_norm = pre_norm
|
413 |
+
|
414 |
+
self.residual_attn = residual_attn
|
415 |
+
self.cross_residual_attn = cross_residual_attn
|
416 |
+
|
417 |
+
norm_class = ScaleNorm if use_scalenorm else nn.LayerNorm
|
418 |
+
norm_class = RMSNorm if use_rmsnorm else norm_class
|
419 |
+
norm_fn = partial(norm_class, dim)
|
420 |
+
|
421 |
+
norm_fn = nn.Identity if use_rezero else norm_fn
|
422 |
+
branch_fn = Rezero if use_rezero else None
|
423 |
+
|
424 |
+
if cross_attend and not only_cross:
|
425 |
+
default_block = ('a', 'c', 'f')
|
426 |
+
elif cross_attend and only_cross:
|
427 |
+
default_block = ('c', 'f')
|
428 |
+
else:
|
429 |
+
default_block = ('a', 'f')
|
430 |
+
|
431 |
+
if macaron:
|
432 |
+
default_block = ('f',) + default_block
|
433 |
+
|
434 |
+
if exists(custom_layers):
|
435 |
+
layer_types = custom_layers
|
436 |
+
elif exists(par_ratio):
|
437 |
+
par_depth = depth * len(default_block)
|
438 |
+
assert 1 < par_ratio <= par_depth, 'par ratio out of range'
|
439 |
+
default_block = tuple(filter(not_equals('f'), default_block))
|
440 |
+
par_attn = par_depth // par_ratio
|
441 |
+
depth_cut = par_depth * 2 // 3 # 2 / 3 attention layer cutoff suggested by PAR paper
|
442 |
+
par_width = (depth_cut + depth_cut // par_attn) // par_attn
|
443 |
+
assert len(default_block) <= par_width, 'default block is too large for par_ratio'
|
444 |
+
par_block = default_block + ('f',) * (par_width - len(default_block))
|
445 |
+
par_head = par_block * par_attn
|
446 |
+
layer_types = par_head + ('f',) * (par_depth - len(par_head))
|
447 |
+
elif exists(sandwich_coef):
|
448 |
+
assert sandwich_coef > 0 and sandwich_coef <= depth, 'sandwich coefficient should be less than the depth'
|
449 |
+
layer_types = ('a',) * sandwich_coef + default_block * (depth - sandwich_coef) + ('f',) * sandwich_coef
|
450 |
+
else:
|
451 |
+
layer_types = default_block * depth
|
452 |
+
|
453 |
+
self.layer_types = layer_types
|
454 |
+
self.num_attn_layers = len(list(filter(equals('a'), layer_types)))
|
455 |
+
|
456 |
+
for layer_type in self.layer_types:
|
457 |
+
if layer_type == 'a':
|
458 |
+
layer = Attention(dim, heads=heads, causal=causal, **attn_kwargs)
|
459 |
+
elif layer_type == 'c':
|
460 |
+
layer = Attention(dim, heads=heads, **attn_kwargs)
|
461 |
+
elif layer_type == 'f':
|
462 |
+
layer = FeedForward(dim, **ff_kwargs)
|
463 |
+
layer = layer if not macaron else Scale(0.5, layer)
|
464 |
+
else:
|
465 |
+
raise Exception(f'invalid layer type {layer_type}')
|
466 |
+
|
467 |
+
if isinstance(layer, Attention) and exists(branch_fn):
|
468 |
+
layer = branch_fn(layer)
|
469 |
+
|
470 |
+
if gate_residual:
|
471 |
+
residual_fn = GRUGating(dim)
|
472 |
+
else:
|
473 |
+
residual_fn = Residual()
|
474 |
+
|
475 |
+
self.layers.append(nn.ModuleList([
|
476 |
+
norm_fn(),
|
477 |
+
layer,
|
478 |
+
residual_fn
|
479 |
+
]))
|
480 |
+
|
481 |
+
def forward(
|
482 |
+
self,
|
483 |
+
x,
|
484 |
+
context=None,
|
485 |
+
mask=None,
|
486 |
+
context_mask=None,
|
487 |
+
mems=None,
|
488 |
+
return_hiddens=False
|
489 |
+
):
|
490 |
+
hiddens = []
|
491 |
+
intermediates = []
|
492 |
+
prev_attn = None
|
493 |
+
prev_cross_attn = None
|
494 |
+
|
495 |
+
mems = mems.copy() if exists(mems) else [None] * self.num_attn_layers
|
496 |
+
|
497 |
+
for ind, (layer_type, (norm, block, residual_fn)) in enumerate(zip(self.layer_types, self.layers)):
|
498 |
+
is_last = ind == (len(self.layers) - 1)
|
499 |
+
|
500 |
+
if layer_type == 'a':
|
501 |
+
hiddens.append(x)
|
502 |
+
layer_mem = mems.pop(0)
|
503 |
+
|
504 |
+
residual = x
|
505 |
+
|
506 |
+
if self.pre_norm:
|
507 |
+
x = norm(x)
|
508 |
+
|
509 |
+
if layer_type == 'a':
|
510 |
+
out, inter = block(x, mask=mask, sinusoidal_emb=self.pia_pos_emb, rel_pos=self.rel_pos,
|
511 |
+
prev_attn=prev_attn, mem=layer_mem)
|
512 |
+
elif layer_type == 'c':
|
513 |
+
out, inter = block(x, context=context, mask=mask, context_mask=context_mask, prev_attn=prev_cross_attn)
|
514 |
+
elif layer_type == 'f':
|
515 |
+
out = block(x)
|
516 |
+
|
517 |
+
x = residual_fn(out, residual)
|
518 |
+
|
519 |
+
if layer_type in ('a', 'c'):
|
520 |
+
intermediates.append(inter)
|
521 |
+
|
522 |
+
if layer_type == 'a' and self.residual_attn:
|
523 |
+
prev_attn = inter.pre_softmax_attn
|
524 |
+
elif layer_type == 'c' and self.cross_residual_attn:
|
525 |
+
prev_cross_attn = inter.pre_softmax_attn
|
526 |
+
|
527 |
+
if not self.pre_norm and not is_last:
|
528 |
+
x = norm(x)
|
529 |
+
|
530 |
+
if return_hiddens:
|
531 |
+
intermediates = LayerIntermediates(
|
532 |
+
hiddens=hiddens,
|
533 |
+
attn_intermediates=intermediates
|
534 |
+
)
|
535 |
+
|
536 |
+
return x, intermediates
|
537 |
+
|
538 |
+
return x
|
539 |
+
|
540 |
+
|
541 |
+
class Encoder(AttentionLayers):
|
542 |
+
def __init__(self, **kwargs):
|
543 |
+
assert 'causal' not in kwargs, 'cannot set causality on encoder'
|
544 |
+
super().__init__(causal=False, **kwargs)
|
545 |
+
|
546 |
+
|
547 |
+
|
548 |
+
class TransformerWrapper(nn.Module):
|
549 |
+
def __init__(
|
550 |
+
self,
|
551 |
+
*,
|
552 |
+
num_tokens,
|
553 |
+
max_seq_len,
|
554 |
+
attn_layers,
|
555 |
+
emb_dim=None,
|
556 |
+
max_mem_len=0.,
|
557 |
+
emb_dropout=0.,
|
558 |
+
num_memory_tokens=None,
|
559 |
+
tie_embedding=False,
|
560 |
+
use_pos_emb=True
|
561 |
+
):
|
562 |
+
super().__init__()
|
563 |
+
assert isinstance(attn_layers, AttentionLayers), 'attention layers must be one of Encoder or Decoder'
|
564 |
+
|
565 |
+
dim = attn_layers.dim
|
566 |
+
emb_dim = default(emb_dim, dim)
|
567 |
+
|
568 |
+
self.max_seq_len = max_seq_len
|
569 |
+
self.max_mem_len = max_mem_len
|
570 |
+
self.num_tokens = num_tokens
|
571 |
+
|
572 |
+
self.token_emb = nn.Embedding(num_tokens, emb_dim)
|
573 |
+
self.pos_emb = AbsolutePositionalEmbedding(emb_dim, max_seq_len) if (
|
574 |
+
use_pos_emb and not attn_layers.has_pos_emb) else always(0)
|
575 |
+
self.emb_dropout = nn.Dropout(emb_dropout)
|
576 |
+
|
577 |
+
self.project_emb = nn.Linear(emb_dim, dim) if emb_dim != dim else nn.Identity()
|
578 |
+
self.attn_layers = attn_layers
|
579 |
+
self.norm = nn.LayerNorm(dim)
|
580 |
+
|
581 |
+
self.init_()
|
582 |
+
|
583 |
+
self.to_logits = nn.Linear(dim, num_tokens) if not tie_embedding else lambda t: t @ self.token_emb.weight.t()
|
584 |
+
|
585 |
+
# memory tokens (like [cls]) from Memory Transformers paper
|
586 |
+
num_memory_tokens = default(num_memory_tokens, 0)
|
587 |
+
self.num_memory_tokens = num_memory_tokens
|
588 |
+
if num_memory_tokens > 0:
|
589 |
+
self.memory_tokens = nn.Parameter(torch.randn(num_memory_tokens, dim))
|
590 |
+
|
591 |
+
# let funnel encoder know number of memory tokens, if specified
|
592 |
+
if hasattr(attn_layers, 'num_memory_tokens'):
|
593 |
+
attn_layers.num_memory_tokens = num_memory_tokens
|
594 |
+
|
595 |
+
def init_(self):
|
596 |
+
nn.init.normal_(self.token_emb.weight, std=0.02)
|
597 |
+
|
598 |
+
def forward(
|
599 |
+
self,
|
600 |
+
x,
|
601 |
+
return_embeddings=False,
|
602 |
+
mask=None,
|
603 |
+
return_mems=False,
|
604 |
+
return_attn=False,
|
605 |
+
mems=None,
|
606 |
+
**kwargs
|
607 |
+
):
|
608 |
+
b, n, device, num_mem = *x.shape, x.device, self.num_memory_tokens
|
609 |
+
x = self.token_emb(x)
|
610 |
+
x += self.pos_emb(x)
|
611 |
+
x = self.emb_dropout(x)
|
612 |
+
|
613 |
+
x = self.project_emb(x)
|
614 |
+
|
615 |
+
if num_mem > 0:
|
616 |
+
mem = repeat(self.memory_tokens, 'n d -> b n d', b=b)
|
617 |
+
x = torch.cat((mem, x), dim=1)
|
618 |
+
|
619 |
+
# auto-handle masking after appending memory tokens
|
620 |
+
if exists(mask):
|
621 |
+
mask = F.pad(mask, (num_mem, 0), value=True)
|
622 |
+
|
623 |
+
x, intermediates = self.attn_layers(x, mask=mask, mems=mems, return_hiddens=True, **kwargs)
|
624 |
+
x = self.norm(x)
|
625 |
+
|
626 |
+
mem, x = x[:, :num_mem], x[:, num_mem:]
|
627 |
+
|
628 |
+
out = self.to_logits(x) if not return_embeddings else x
|
629 |
+
|
630 |
+
if return_mems:
|
631 |
+
hiddens = intermediates.hiddens
|
632 |
+
new_mems = list(map(lambda pair: torch.cat(pair, dim=-2), zip(mems, hiddens))) if exists(mems) else hiddens
|
633 |
+
new_mems = list(map(lambda t: t[..., -self.max_mem_len:, :].detach(), new_mems))
|
634 |
+
return out, new_mems
|
635 |
+
|
636 |
+
if return_attn:
|
637 |
+
attn_maps = list(map(lambda t: t.post_softmax_attn, intermediates.attn_intermediates))
|
638 |
+
return out, attn_maps
|
639 |
+
|
640 |
+
return out
|
641 |
+
|
ldm/util.py
ADDED
@@ -0,0 +1,211 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import importlib
|
2 |
+
|
3 |
+
import torch
|
4 |
+
import numpy as np
|
5 |
+
from collections import abc
|
6 |
+
import os
|
7 |
+
import random
|
8 |
+
|
9 |
+
import multiprocessing as mp
|
10 |
+
from threading import Thread
|
11 |
+
from queue import Queue
|
12 |
+
|
13 |
+
from inspect import isfunction
|
14 |
+
from PIL import Image, ImageDraw, ImageFont
|
15 |
+
|
16 |
+
|
17 |
+
def log_txt_as_img(wh, xc, size=10):
|
18 |
+
# wh a tuple of (width, height)
|
19 |
+
# xc a list of captions to plot
|
20 |
+
b = len(xc)
|
21 |
+
txts = list()
|
22 |
+
for bi in range(b):
|
23 |
+
txt = Image.new("RGB", wh, color="white")
|
24 |
+
draw = ImageDraw.Draw(txt)
|
25 |
+
font = ImageFont.truetype('data/DejaVuSans.ttf', size=size)
|
26 |
+
nc = int(40 * (wh[0] / 256))
|
27 |
+
lines = "\n".join(xc[bi][start:start + nc] for start in range(0, len(xc[bi]), nc))
|
28 |
+
|
29 |
+
try:
|
30 |
+
draw.text((0, 0), lines, fill="black", font=font)
|
31 |
+
except UnicodeEncodeError:
|
32 |
+
print("Cant encode string for logging. Skipping.")
|
33 |
+
|
34 |
+
txt = np.array(txt).transpose(2, 0, 1) / 127.5 - 1.0
|
35 |
+
txts.append(txt)
|
36 |
+
txts = np.stack(txts)
|
37 |
+
txts = torch.tensor(txts)
|
38 |
+
return txts
|
39 |
+
|
40 |
+
|
41 |
+
def ismap(x):
|
42 |
+
if not isinstance(x, torch.Tensor):
|
43 |
+
return False
|
44 |
+
return (len(x.shape) == 4) and (x.shape[1] > 3)
|
45 |
+
|
46 |
+
|
47 |
+
def isimage(x):
|
48 |
+
if not isinstance(x, torch.Tensor):
|
49 |
+
return False
|
50 |
+
return (len(x.shape) == 4) and (x.shape[1] == 3 or x.shape[1] == 1)
|
51 |
+
|
52 |
+
|
53 |
+
def exists(x):
|
54 |
+
return x is not None
|
55 |
+
|
56 |
+
|
57 |
+
def default(val, d):
|
58 |
+
if exists(val):
|
59 |
+
return val
|
60 |
+
return d() if isfunction(d) else d
|
61 |
+
|
62 |
+
|
63 |
+
def mean_flat(tensor):
|
64 |
+
"""
|
65 |
+
https://github.com/openai/guided-diffusion/blob/27c20a8fab9cb472df5d6bdd6c8d11c8f430b924/guided_diffusion/nn.py#L86
|
66 |
+
Take the mean over all non-batch dimensions.
|
67 |
+
"""
|
68 |
+
return tensor.mean(dim=list(range(1, len(tensor.shape))))
|
69 |
+
|
70 |
+
|
71 |
+
def count_params(model, verbose=False):
|
72 |
+
total_params = sum(p.numel() for p in model.parameters())
|
73 |
+
if verbose:
|
74 |
+
print(f"{model.__class__.__name__} has {total_params * 1.e-6:.2f} M params.")
|
75 |
+
return total_params
|
76 |
+
|
77 |
+
|
78 |
+
def instantiate_from_config(config):
|
79 |
+
if not "target" in config:
|
80 |
+
if config == '__is_first_stage__':
|
81 |
+
return None
|
82 |
+
elif config == "__is_unconditional__":
|
83 |
+
return None
|
84 |
+
raise KeyError("Expected key `target` to instantiate.")
|
85 |
+
return get_obj_from_str(config["target"])(**config.get("params", dict()))
|
86 |
+
|
87 |
+
|
88 |
+
def get_obj_from_str(string, reload=False):
|
89 |
+
module, cls = string.rsplit(".", 1)
|
90 |
+
if reload:
|
91 |
+
module_imp = importlib.import_module(module)
|
92 |
+
importlib.reload(module_imp)
|
93 |
+
return getattr(importlib.import_module(module, package=None), cls)
|
94 |
+
|
95 |
+
|
96 |
+
def _do_parallel_data_prefetch(func, Q, data, idx, idx_to_fn=False):
|
97 |
+
# create dummy dataset instance
|
98 |
+
|
99 |
+
# run prefetching
|
100 |
+
if idx_to_fn:
|
101 |
+
res = func(data, worker_id=idx)
|
102 |
+
else:
|
103 |
+
res = func(data)
|
104 |
+
Q.put([idx, res])
|
105 |
+
Q.put("Done")
|
106 |
+
|
107 |
+
|
108 |
+
def parallel_data_prefetch(
|
109 |
+
func: callable, data, n_proc, target_data_type="ndarray", cpu_intensive=True, use_worker_id=False
|
110 |
+
):
|
111 |
+
# if target_data_type not in ["ndarray", "list"]:
|
112 |
+
# raise ValueError(
|
113 |
+
# "Data, which is passed to parallel_data_prefetch has to be either of type list or ndarray."
|
114 |
+
# )
|
115 |
+
if isinstance(data, np.ndarray) and target_data_type == "list":
|
116 |
+
raise ValueError("list expected but function got ndarray.")
|
117 |
+
elif isinstance(data, abc.Iterable):
|
118 |
+
if isinstance(data, dict):
|
119 |
+
print(
|
120 |
+
f'WARNING:"data" argument passed to parallel_data_prefetch is a dict: Using only its values and disregarding keys.'
|
121 |
+
)
|
122 |
+
data = list(data.values())
|
123 |
+
if target_data_type == "ndarray":
|
124 |
+
data = np.asarray(data)
|
125 |
+
else:
|
126 |
+
data = list(data)
|
127 |
+
else:
|
128 |
+
raise TypeError(
|
129 |
+
f"The data, that shall be processed parallel has to be either an np.ndarray or an Iterable, but is actually {type(data)}."
|
130 |
+
)
|
131 |
+
|
132 |
+
if cpu_intensive:
|
133 |
+
Q = mp.Queue(1000)
|
134 |
+
proc = mp.Process
|
135 |
+
else:
|
136 |
+
Q = Queue(1000)
|
137 |
+
proc = Thread
|
138 |
+
# spawn processes
|
139 |
+
if target_data_type == "ndarray":
|
140 |
+
arguments = [
|
141 |
+
[func, Q, part, i, use_worker_id]
|
142 |
+
for i, part in enumerate(np.array_split(data, n_proc))
|
143 |
+
]
|
144 |
+
else:
|
145 |
+
step = (
|
146 |
+
int(len(data) / n_proc + 1)
|
147 |
+
if len(data) % n_proc != 0
|
148 |
+
else int(len(data) / n_proc)
|
149 |
+
)
|
150 |
+
arguments = [
|
151 |
+
[func, Q, part, i, use_worker_id]
|
152 |
+
for i, part in enumerate(
|
153 |
+
[data[i: i + step] for i in range(0, len(data), step)]
|
154 |
+
)
|
155 |
+
]
|
156 |
+
processes = []
|
157 |
+
for i in range(n_proc):
|
158 |
+
p = proc(target=_do_parallel_data_prefetch, args=arguments[i])
|
159 |
+
processes += [p]
|
160 |
+
|
161 |
+
# start processes
|
162 |
+
print(f"Start prefetching...")
|
163 |
+
import time
|
164 |
+
|
165 |
+
start = time.time()
|
166 |
+
gather_res = [[] for _ in range(n_proc)]
|
167 |
+
try:
|
168 |
+
for p in processes:
|
169 |
+
p.start()
|
170 |
+
|
171 |
+
k = 0
|
172 |
+
while k < n_proc:
|
173 |
+
# get result
|
174 |
+
res = Q.get()
|
175 |
+
if res == "Done":
|
176 |
+
k += 1
|
177 |
+
else:
|
178 |
+
gather_res[res[0]] = res[1]
|
179 |
+
|
180 |
+
except Exception as e:
|
181 |
+
print("Exception: ", e)
|
182 |
+
for p in processes:
|
183 |
+
p.terminate()
|
184 |
+
|
185 |
+
raise e
|
186 |
+
finally:
|
187 |
+
for p in processes:
|
188 |
+
p.join()
|
189 |
+
print(f"Prefetching complete. [{time.time() - start} sec.]")
|
190 |
+
|
191 |
+
if target_data_type == 'ndarray':
|
192 |
+
if not isinstance(gather_res[0], np.ndarray):
|
193 |
+
return np.concatenate([np.asarray(r) for r in gather_res], axis=0)
|
194 |
+
|
195 |
+
# order outputs
|
196 |
+
return np.concatenate(gather_res, axis=0)
|
197 |
+
elif target_data_type == 'list':
|
198 |
+
out = []
|
199 |
+
for r in gather_res:
|
200 |
+
out.extend(r)
|
201 |
+
return out
|
202 |
+
else:
|
203 |
+
return gather_res
|
204 |
+
|
205 |
+
def seed_everything(seed: int=0):
|
206 |
+
random.seed(seed)
|
207 |
+
os.environ["PYTHONHASHSEED"] = str(seed)
|
208 |
+
np.random.seed(seed)
|
209 |
+
torch.manual_seed(seed)
|
210 |
+
torch.cuda.manual_seed(seed)
|
211 |
+
torch.backends.cudnn.deterministic = True
|
requirements.txt
CHANGED
@@ -9,5 +9,6 @@ kornia==0.6.11
|
|
9 |
transformers==4.27.4
|
10 |
dill==0.3.6
|
11 |
gradio==3.26.0
|
|
|
12 |
gdown==4.7.1
|
13 |
-
|
|
|
9 |
transformers==4.27.4
|
10 |
dill==0.3.6
|
11 |
gradio==3.26.0
|
12 |
+
torchmetrics==0.11.4
|
13 |
gdown==4.7.1
|
14 |
+
omegaconf==2.3.0
|
utils.py
CHANGED
@@ -1,25 +1,13 @@
|
|
1 |
-
import logging
|
2 |
-
import os
|
3 |
-
import random
|
4 |
-
import tarfile
|
5 |
from typing import Tuple
|
6 |
-
|
7 |
-
import dill
|
8 |
-
import gdown
|
9 |
-
import numpy as np
|
10 |
-
import torch
|
11 |
from PIL import Image
|
12 |
from torchvision.transforms import ToTensor
|
13 |
|
14 |
-
logger = logging.getLogger(__file__)
|
15 |
-
|
16 |
to_tensor = ToTensor()
|
17 |
|
18 |
-
|
19 |
def preprocess_image(
|
20 |
image: Image, resize_shape: Tuple[int, int] = (256, 256), center_crop=True
|
21 |
):
|
22 |
-
|
23 |
|
24 |
if center_crop:
|
25 |
width, height = image.size
|
@@ -30,49 +18,11 @@ def preprocess_image(
|
|
30 |
right = (width + crop_size) // 2
|
31 |
bottom = (height + crop_size) // 2
|
32 |
|
33 |
-
|
34 |
-
|
35 |
-
processed_image = processed_image.resize(resize_shape)
|
36 |
-
|
37 |
-
image = to_tensor(processed_image)
|
38 |
-
image = image.unsqueeze(0) * 2 - 1
|
39 |
-
|
40 |
-
return processed_image, image
|
41 |
-
|
42 |
-
|
43 |
-
def download_artifacts(output_path: str):
|
44 |
-
logger.error("Downloading the model artifacts...")
|
45 |
-
if not os.path.exists(output_path):
|
46 |
-
gdown.download(id=os.environ["GDRIVE_ID"], output=output_path, quiet=True)
|
47 |
-
|
48 |
-
|
49 |
-
def extract_artifacts(path: str):
|
50 |
-
logger.error("Extracting the model artifacts...")
|
51 |
-
if not os.path.exists("model.pkl"):
|
52 |
-
with tarfile.open(path) as tar:
|
53 |
-
tar.extractall()
|
54 |
-
|
55 |
-
|
56 |
-
def setup_environment():
|
57 |
-
os.environ["PYTHONPATH"] = os.getcwd()
|
58 |
-
|
59 |
-
artifacts_path = "artifacts.tar.gz"
|
60 |
-
|
61 |
-
download_artifacts(output_path=artifacts_path)
|
62 |
-
|
63 |
-
extract_artifacts(path=artifacts_path)
|
64 |
-
|
65 |
-
|
66 |
-
def get_predictor():
|
67 |
-
logger.error("Loading the predictor...")
|
68 |
-
with open("model.pkl", "rb") as fp:
|
69 |
-
return dill.load(fp)
|
70 |
|
|
|
|
|
|
|
|
|
71 |
|
72 |
-
|
73 |
-
random.seed(seed)
|
74 |
-
os.environ["PYTHONHASHSEED"] = str(seed)
|
75 |
-
np.random.seed(seed)
|
76 |
-
torch.manual_seed(seed)
|
77 |
-
torch.cuda.manual_seed(seed)
|
78 |
-
torch.backends.cudnn.deterministic = True
|
|
|
|
|
|
|
|
|
|
|
1 |
from typing import Tuple
|
|
|
|
|
|
|
|
|
|
|
2 |
from PIL import Image
|
3 |
from torchvision.transforms import ToTensor
|
4 |
|
|
|
|
|
5 |
to_tensor = ToTensor()
|
6 |
|
|
|
7 |
def preprocess_image(
|
8 |
image: Image, resize_shape: Tuple[int, int] = (256, 256), center_crop=True
|
9 |
):
|
10 |
+
pil_image = image
|
11 |
|
12 |
if center_crop:
|
13 |
width, height = image.size
|
|
|
18 |
right = (width + crop_size) // 2
|
19 |
bottom = (height + crop_size) // 2
|
20 |
|
21 |
+
pil_image = image.crop((left, top, right, bottom))
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
22 |
|
23 |
+
pil_image = pil_image.resize(resize_shape)
|
24 |
+
|
25 |
+
tensor_image = to_tensor(pil_image)
|
26 |
+
tensor_image = tensor_image.unsqueeze(0) * 2 - 1
|
27 |
|
28 |
+
return pil_image, tensor_image
|
|
|
|
|
|
|
|
|
|
|
|