Logan Zoellner
commited on
Commit
·
bfc97b7
1
Parent(s):
73630d0
inital commit
Browse files- app.py +213 -0
- requirements.txt +8 -0
app.py
ADDED
@@ -0,0 +1,213 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
|
2 |
+
|
3 |
+
#%%capture
|
4 |
+
#!git lfs install
|
5 |
+
#!git clone https://huggingface.co/Cene655/ImagenT5-3B
|
6 |
+
|
7 |
+
#%%capture
|
8 |
+
#!pip install git+https://github.com/cene555/Imagen-pytorch.git
|
9 |
+
#!pip install git+https://github.com/openai/CLIP.git
|
10 |
+
|
11 |
+
#%%capture
|
12 |
+
#!git clone https://github.com/xinntao/Real-ESRGAN.git
|
13 |
+
|
14 |
+
#%cd Real-ESRGAN
|
15 |
+
|
16 |
+
#%%capture
|
17 |
+
#!pip install basicsr
|
18 |
+
# facexlib and gfpgan are for face enhancement
|
19 |
+
#!pip install facexlib
|
20 |
+
#!pip install gfpgan
|
21 |
+
|
22 |
+
#%%capture
|
23 |
+
#!pip install -r requirements.txt
|
24 |
+
#!python setup.py develop
|
25 |
+
#!wget https://github.com/xinntao/Real-ESRGAN/releases/download/v0.1.0/RealESRGAN_x4plus.pth -P experiments/pretrained_models
|
26 |
+
|
27 |
+
#Imports
|
28 |
+
|
29 |
+
from PIL import Image
|
30 |
+
from IPython.display import display
|
31 |
+
import torch as th
|
32 |
+
from imagen_pytorch.model_creation import create_model_and_diffusion as create_model_and_diffusion_dalle2
|
33 |
+
from imagen_pytorch.model_creation import model_and_diffusion_defaults as model_and_diffusion_defaults_dalle2
|
34 |
+
from transformers import AutoTokenizer
|
35 |
+
import cv2
|
36 |
+
|
37 |
+
import glob
|
38 |
+
import os
|
39 |
+
from basicsr.archs.rrdbnet_arch import RRDBNet
|
40 |
+
from realesrgan import RealESRGANer
|
41 |
+
from realesrgan.archs.srvgg_arch import SRVGGNetCompact
|
42 |
+
from gfpgan import GFPGANer
|
43 |
+
|
44 |
+
has_cuda = th.cuda.is_available()
|
45 |
+
device = th.device('cpu' if not has_cuda else 'cuda')
|
46 |
+
|
47 |
+
Setting Up
|
48 |
+
|
49 |
+
def model_fn(x_t, ts, **kwargs):
|
50 |
+
guidance_scale = 5
|
51 |
+
half = x_t[: len(x_t) // 2]
|
52 |
+
combined = th.cat([half, half], dim=0)
|
53 |
+
model_out = model(combined, ts, **kwargs)
|
54 |
+
eps, rest = model_out[:, :3], model_out[:, 3:]
|
55 |
+
cond_eps, uncond_eps = th.split(eps, len(eps) // 2, dim=0)
|
56 |
+
half_eps = uncond_eps + guidance_scale * (cond_eps - uncond_eps)
|
57 |
+
eps = th.cat([half_eps, half_eps], dim=0)
|
58 |
+
return th.cat([eps, rest], dim=1)
|
59 |
+
|
60 |
+
def show_images(batch: th.Tensor):
|
61 |
+
""" Display a batch of images inline."""
|
62 |
+
scaled = ((batch + 1)*127.5).round().clamp(0,255).to(th.uint8).cpu()
|
63 |
+
reshaped = scaled.permute(2, 0, 3, 1).reshape([batch.shape[2], -1, 3])
|
64 |
+
display(Image.fromarray(reshaped.numpy()))
|
65 |
+
|
66 |
+
def get_numpy_img(img):
|
67 |
+
scaled = ((img + 1)*127.5).round().clamp(0,255).to(th.uint8).cpu()
|
68 |
+
reshaped = scaled.permute(2, 0, 3, 1).reshape([img.shape[2], -1, 3])
|
69 |
+
return cv2.cvtColor(reshaped.numpy(), cv2.COLOR_BGR2RGB)
|
70 |
+
|
71 |
+
def _fix_path(path):
|
72 |
+
d = th.load(path)
|
73 |
+
checkpoint = {}
|
74 |
+
for key in d.keys():
|
75 |
+
checkpoint[key.replace('module.','')] = d[key]
|
76 |
+
return checkpoint
|
77 |
+
|
78 |
+
options = model_and_diffusion_defaults_dalle2()
|
79 |
+
options['use_fp16'] = False
|
80 |
+
options['diffusion_steps'] = 200
|
81 |
+
options['num_res_blocks'] = 3
|
82 |
+
options['t5_name'] = 't5-3b'
|
83 |
+
options['cache_text_emb'] = True
|
84 |
+
model, diffusion = create_model_and_diffusion_dalle2(**options)
|
85 |
+
|
86 |
+
model.eval()
|
87 |
+
|
88 |
+
#if has_cuda:
|
89 |
+
# model.convert_to_fp16()
|
90 |
+
|
91 |
+
model.to(device)
|
92 |
+
|
93 |
+
model.load_state_dict(_fix_path('/content/ImagenT5-3B/model.pt'))
|
94 |
+
print('total base parameters', sum(x.numel() for x in model.parameters()))
|
95 |
+
|
96 |
+
total base parameters 1550556742
|
97 |
+
|
98 |
+
num_params = sum(param.numel() for param in model.parameters())
|
99 |
+
num_params
|
100 |
+
|
101 |
+
1550556742
|
102 |
+
|
103 |
+
realesrgan_model = RRDBNet(num_in_ch=3, num_out_ch=3, num_feat=64,
|
104 |
+
num_block=23, num_grow_ch=32, scale=4)
|
105 |
+
|
106 |
+
netscale = 4
|
107 |
+
|
108 |
+
upsampler = RealESRGANer(
|
109 |
+
scale=netscale,
|
110 |
+
model_path='/content/Real-ESRGAN/experiments/pretrained_models/RealESRGAN_x4plus.pth',
|
111 |
+
model=realesrgan_model,
|
112 |
+
tile=0,
|
113 |
+
tile_pad=10,
|
114 |
+
pre_pad=0,
|
115 |
+
half=True
|
116 |
+
)
|
117 |
+
|
118 |
+
face_enhancer = GFPGANer(
|
119 |
+
model_path='https://github.com/TencentARC/GFPGAN/releases/download/v1.3.0/GFPGANv1.3.pth',
|
120 |
+
upscale=4,
|
121 |
+
arch='clean',
|
122 |
+
channel_multiplier=2,
|
123 |
+
bg_upsampler=upsampler
|
124 |
+
)
|
125 |
+
|
126 |
+
tokenizer = AutoTokenizer.from_pretrained(options['t5_name'])
|
127 |
+
|
128 |
+
/usr/local/lib/python3.7/dist-packages/transformers/models/t5/tokenization_t5_fast.py:161: FutureWarning: This tokenizer was incorrectly instantiated with a model max length of 512 which will be corrected in Transformers v5.
|
129 |
+
For now, this behavior is kept to avoid breaking backwards compatibility when padding/encoding with `truncation is True`.
|
130 |
+
- Be aware that you SHOULD NOT rely on t5-3b automatically truncating your input to 512 when padding/encoding.
|
131 |
+
- If you want to encode/pad to sequences longer than 512 you can either instantiate this tokenizer with `model_max_length` or pass `max_length` when encoding/padding.
|
132 |
+
- To avoid this warning, please instantiate this tokenizer with `model_max_length` set to your preferred value.
|
133 |
+
FutureWarning,
|
134 |
+
|
135 |
+
#@title What do you want to generate?
|
136 |
+
|
137 |
+
prompt = 'A photo of cat'#@param {type:"string"}
|
138 |
+
|
139 |
+
def gen_img(prompt):
|
140 |
+
|
141 |
+
text_encoding = tokenizer(
|
142 |
+
prompt,
|
143 |
+
max_length=128,
|
144 |
+
padding="max_length",
|
145 |
+
truncation=True,
|
146 |
+
return_attention_mask=True,
|
147 |
+
add_special_tokens=True,
|
148 |
+
return_tensors="pt"
|
149 |
+
)
|
150 |
+
|
151 |
+
uncond_text_encoding = tokenizer(
|
152 |
+
'',
|
153 |
+
max_length=128,
|
154 |
+
padding="max_length",
|
155 |
+
truncation=True,
|
156 |
+
return_attention_mask=True,
|
157 |
+
add_special_tokens=True,
|
158 |
+
return_tensors="pt"
|
159 |
+
)
|
160 |
+
|
161 |
+
import numpy as np
|
162 |
+
batch_size = 4
|
163 |
+
cond_tokens = th.from_numpy(np.array([text_encoding['input_ids'][0].numpy() for i in range(batch_size)]))
|
164 |
+
uncond_tokens = th.from_numpy(np.array([uncond_text_encoding['input_ids'][0].numpy() for i in range(batch_size)]))
|
165 |
+
cond_attention_mask = th.from_numpy(np.array([text_encoding['attention_mask'][0].numpy() for i in range(batch_size)]))
|
166 |
+
uncond_attention_mask = th.from_numpy(np.array([uncond_text_encoding['attention_mask'][0].numpy() for i in range(batch_size)]))
|
167 |
+
model_kwargs = {}
|
168 |
+
model_kwargs["tokens"] = th.cat((cond_tokens,
|
169 |
+
uncond_tokens)).to(device)
|
170 |
+
model_kwargs["mask"] = th.cat((cond_attention_mask,
|
171 |
+
uncond_attention_mask)).to(device)
|
172 |
+
|
173 |
+
Generation
|
174 |
+
|
175 |
+
model.del_cache()
|
176 |
+
sample = diffusion.p_sample_loop(
|
177 |
+
model_fn,
|
178 |
+
(batch_size * 2, 3, 64, 64),
|
179 |
+
clip_denoised=True,
|
180 |
+
model_kwargs=model_kwargs,
|
181 |
+
device='cuda',
|
182 |
+
progress=True,
|
183 |
+
)[:batch_size]
|
184 |
+
model.del_cache()
|
185 |
+
|
186 |
+
return sample
|
187 |
+
|
188 |
+
demo = gr.Blocks()
|
189 |
+
|
190 |
+
with demo:
|
191 |
+
gr.Markdown("<h1><center>cene555/Imagen-pytorch</center></h1>")
|
192 |
+
gr.Markdown(
|
193 |
+
"<div>github repo <a href='https://github.com/cene555/Imagen-pytorch/blob/main/images/2.jpg'>here</a></div>"
|
194 |
+
"<div>hf model <a href='https://huggingface.co/Cene655/ImagenT5-3B/tree/main'>here</a></div>"
|
195 |
+
)
|
196 |
+
|
197 |
+
with gr.Row():
|
198 |
+
b0 = gr.Button("generate")
|
199 |
+
b1 = gr.Button("upscale")
|
200 |
+
|
201 |
+
with gr.Row():
|
202 |
+
desc = gr.Textbox(label="description",placeholder="an impressionist painting of a white vase")
|
203 |
+
|
204 |
+
with gr.Row():
|
205 |
+
intermediate_image = gr.Image(label="portrait",type="filepath", shape=(256,256))
|
206 |
+
output_image = gr.Image(label="portrait",type="filepath", shape=(256,256))
|
207 |
+
|
208 |
+
b0.click(gen_img,inputs=[desc],outputs=[intermediate_image])
|
209 |
+
b1.click(upscale_img, inputs=[ intermediate_image], outputs=output_image)
|
210 |
+
#examples=examples
|
211 |
+
|
212 |
+
demo.launch(enable_queue=True, debug=True)
|
213 |
+
|
requirements.txt
ADDED
@@ -0,0 +1,8 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
git+https://huggingface.co/Cene655/ImagenT5-3B
|
2 |
+
git+https://github.com/cene555/Imagen-pytorch.git
|
3 |
+
git+https://github.com/openai/CLIP.git
|
4 |
+
git+https://github.com/xinntao/Real-ESRGAN.git
|
5 |
+
basicsr
|
6 |
+
facexlib
|
7 |
+
gfpgan
|
8 |
+
|