File size: 1,427 Bytes
9ebe755 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 |
from __future__ import annotations
import torch
import torch.nn as nn
from monai.utils import optional_import
from torch.cuda.amp import autocast
tqdm, has_tqdm = optional_import("tqdm", name="tqdm")
class Sampler:
def __init__(self) -> None:
super().__init__()
@torch.no_grad()
def sampling_fn(
self,
input_noise: torch.Tensor,
autoencoder_model: nn.Module,
diffusion_model: nn.Module,
scheduler: nn.Module,
conditioning: torch.Tensor,
) -> torch.Tensor:
if has_tqdm:
progress_bar = tqdm(scheduler.timesteps)
else:
progress_bar = iter(scheduler.timesteps)
image = input_noise
cond_concat = conditioning.squeeze(1).unsqueeze(-1).unsqueeze(-1).unsqueeze(-1)
cond_concat = cond_concat.expand(list(cond_concat.shape[0:2]) + list(input_noise.shape[2:]))
for t in progress_bar:
with torch.no_grad():
model_output = diffusion_model(
torch.cat((image, cond_concat), dim=1),
timesteps=torch.Tensor((t,)).to(input_noise.device).long(),
context=conditioning,
)
image, _ = scheduler.step(model_output, t, image)
with torch.no_grad():
with autocast():
sample = autoencoder_model.decode_stage_2_outputs(image)
return sample
|