pipeline_ddpm.py 1.86 KB
Newer Older
Patrick von Platen's avatar
Patrick von Platen committed
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
# Copyright 2022 The HuggingFace Team. All rights reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and

# limitations under the License.


import torch

anton-l's avatar
anton-l committed
19
from tqdm.auto import tqdm
Patrick von Platen's avatar
Patrick von Platen committed
20

21
from ...pipeline_utils import DiffusionPipeline
Patrick von Platen's avatar
Patrick von Platen committed
22
23


Patrick von Platen's avatar
Patrick von Platen committed
24
class DDPMPipeline(DiffusionPipeline):
25
    def __init__(self, unet, scheduler):
Patrick von Platen's avatar
Patrick von Platen committed
26
        super().__init__()
27
28
        scheduler = scheduler.set_format("pt")
        self.register_modules(unet=unet, scheduler=scheduler)
Patrick von Platen's avatar
Patrick von Platen committed
29

Patrick von Platen's avatar
Patrick von Platen committed
30
    @torch.no_grad()
Patrick von Platen's avatar
Patrick von Platen committed
31
32
33
34
35
36
37
    def __call__(self, batch_size=1, generator=None, torch_device=None):
        if torch_device is None:
            torch_device = "cuda" if torch.cuda.is_available() else "cpu"

        self.unet.to(torch_device)

        # Sample gaussian noise to begin loop
Patrick von Platen's avatar
Patrick von Platen committed
38
        image = torch.randn(
Patrick von Platen's avatar
Patrick von Platen committed
39
            (batch_size, self.unet.in_channels, self.unet.image_size, self.unet.image_size),
Patrick von Platen's avatar
Patrick von Platen committed
40
41
            generator=generator,
        )
Patrick von Platen's avatar
Patrick von Platen committed
42
        image = image.to(torch_device)
Patrick von Platen's avatar
Patrick von Platen committed
43

44
45
46
47
        # set step values
        self.scheduler.set_timesteps(1000)

        for t in tqdm(self.scheduler.timesteps):
Patrick von Platen's avatar
Patrick von Platen committed
48
            # 1. predict noise model_output
Patrick von Platen's avatar
Patrick von Platen committed
49
            model_output = self.unet(image, t)["sample"]
Patrick von Platen's avatar
Patrick von Platen committed
50

Patrick von Platen's avatar
Patrick von Platen committed
51
            # 2. predict previous mean of image x_t-1
Patrick von Platen's avatar
Patrick von Platen committed
52
            pred_prev_image = self.scheduler.step(model_output, t, image)["prev_sample"]
Patrick von Platen's avatar
Patrick von Platen committed
53

Patrick von Platen's avatar
Patrick von Platen committed
54
55
            # 3. set current image to prev_image: x_t -> x_t-1
            image = pred_prev_image
Patrick von Platen's avatar
Patrick von Platen committed
56

Patrick von Platen's avatar
Patrick von Platen committed
57
        return {"sample": image}