File size: 8,381 Bytes
ffead1e
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
# 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.

from typing import List, Optional, Tuple, Union

import PIL
import torch
from torchvision import transforms

from diffusers.pipeline_utils import DiffusionPipeline, ImagePipelineOutput
from diffusers.schedulers import DDIMScheduler
from diffusers.utils import randn_tensor


trans = transforms.Compose(
    [
        transforms.Resize((256, 256)),
        transforms.ToTensor(),
        transforms.Normalize([0.5], [0.5]),
    ]
)


def preprocess(image):
    if isinstance(image, torch.Tensor):
        return image
    elif isinstance(image, PIL.Image.Image):
        image = [image]

    image = [trans(img.convert("RGB")) for img in image]
    image = torch.stack(image)
    return image


class DDIMNoiseComparativeAnalysisPipeline(DiffusionPipeline):
    r"""
    This model inherits from [`DiffusionPipeline`]. Check the superclass documentation for the generic methods the
    library implements for all the pipelines (such as downloading or saving, running on a particular device, etc.)

    Parameters:
        unet ([`UNet2DModel`]): U-Net architecture to denoise the encoded image.
        scheduler ([`SchedulerMixin`]):
            A scheduler to be used in combination with `unet` to denoise the encoded image. Can be one of
            [`DDPMScheduler`], or [`DDIMScheduler`].
    """

    def __init__(self, unet, scheduler):
        super().__init__()

        # make sure scheduler can always be converted to DDIM
        scheduler = DDIMScheduler.from_config(scheduler.config)

        self.register_modules(unet=unet, scheduler=scheduler)

    def check_inputs(self, strength):
        if strength < 0 or strength > 1:
            raise ValueError(f"The value of strength should in [0.0, 1.0] but is {strength}")

    def get_timesteps(self, num_inference_steps, strength, device):
        # get the original timestep using init_timestep
        init_timestep = min(int(num_inference_steps * strength), num_inference_steps)

        t_start = max(num_inference_steps - init_timestep, 0)
        timesteps = self.scheduler.timesteps[t_start:]

        return timesteps, num_inference_steps - t_start

    def prepare_latents(self, image, timestep, batch_size, dtype, device, generator=None):
        if not isinstance(image, (torch.Tensor, PIL.Image.Image, list)):
            raise ValueError(
                f"`image` has to be of type `torch.Tensor`, `PIL.Image.Image` or list but is {type(image)}"
            )

        init_latents = image.to(device=device, dtype=dtype)

        if isinstance(generator, list) and len(generator) != batch_size:
            raise ValueError(
                f"You have passed a list of generators of length {len(generator)}, but requested an effective batch"
                f" size of {batch_size}. Make sure the batch size matches the length of the generators."
            )

        shape = init_latents.shape
        noise = randn_tensor(shape, generator=generator, device=device, dtype=dtype)

        # get latents
        print("add noise to latents at timestep", timestep)
        init_latents = self.scheduler.add_noise(init_latents, noise, timestep)
        latents = init_latents

        return latents

    @torch.no_grad()
    def __call__(
        self,
        image: Union[torch.FloatTensor, PIL.Image.Image] = None,
        strength: float = 0.8,
        batch_size: int = 1,
        generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
        eta: float = 0.0,
        num_inference_steps: int = 50,
        use_clipped_model_output: Optional[bool] = None,
        output_type: Optional[str] = "pil",
        return_dict: bool = True,
    ) -> Union[ImagePipelineOutput, Tuple]:
        r"""
        Args:
            image (`torch.FloatTensor` or `PIL.Image.Image`):
                `Image`, or tensor representing an image batch, that will be used as the starting point for the
                process.
            strength (`float`, *optional*, defaults to 0.8):
                Conceptually, indicates how much to transform the reference `image`. Must be between 0 and 1. `image`
                will be used as a starting point, adding more noise to it the larger the `strength`. The number of
                denoising steps depends on the amount of noise initially added. When `strength` is 1, added noise will
                be maximum and the denoising process will run for the full number of iterations specified in
                `num_inference_steps`. A value of 1, therefore, essentially ignores `image`.
            batch_size (`int`, *optional*, defaults to 1):
                The number of images to generate.
            generator (`torch.Generator`, *optional*):
                One or a list of [torch generator(s)](https://pytorch.org/docs/stable/generated/torch.Generator.html)
                to make generation deterministic.
            eta (`float`, *optional*, defaults to 0.0):
                The eta parameter which controls the scale of the variance (0 is DDIM and 1 is one type of DDPM).
            num_inference_steps (`int`, *optional*, defaults to 50):
                The number of denoising steps. More denoising steps usually lead to a higher quality image at the
                expense of slower inference.
            use_clipped_model_output (`bool`, *optional*, defaults to `None`):
                if `True` or `False`, see documentation for `DDIMScheduler.step`. If `None`, nothing is passed
                downstream to the scheduler. So use `None` for schedulers which don't support this argument.
            output_type (`str`, *optional*, defaults to `"pil"`):
                The output format of the generate image. Choose between
                [PIL](https://pillow.readthedocs.io/en/stable/): `PIL.Image.Image` or `np.array`.
            return_dict (`bool`, *optional*, defaults to `True`):
                Whether or not to return a [`~pipelines.ImagePipelineOutput`] instead of a plain tuple.

        Returns:
            [`~pipelines.ImagePipelineOutput`] or `tuple`: [`~pipelines.utils.ImagePipelineOutput`] if `return_dict` is
            True, otherwise a `tuple. When returning a tuple, the first element is a list with the generated images.
        """
        # 1. Check inputs. Raise error if not correct
        self.check_inputs(strength)

        # 2. Preprocess image
        image = preprocess(image)

        # 3. set timesteps
        self.scheduler.set_timesteps(num_inference_steps, device=self.device)
        timesteps, num_inference_steps = self.get_timesteps(num_inference_steps, strength, self.device)
        latent_timestep = timesteps[:1].repeat(batch_size)

        # 4. Prepare latent variables
        latents = self.prepare_latents(image, latent_timestep, batch_size, self.unet.dtype, self.device, generator)
        image = latents

        # 5. Denoising loop
        for t in self.progress_bar(timesteps):
            # 1. predict noise model_output
            model_output = self.unet(image, t).sample

            # 2. predict previous mean of image x_t-1 and add variance depending on eta
            # eta corresponds to η in paper and should be between [0, 1]
            # do x_t -> x_t-1
            image = self.scheduler.step(
                model_output,
                t,
                image,
                eta=eta,
                use_clipped_model_output=use_clipped_model_output,
                generator=generator,
            ).prev_sample

        image = (image / 2 + 0.5).clamp(0, 1)
        image = image.cpu().permute(0, 2, 3, 1).numpy()
        if output_type == "pil":
            image = self.numpy_to_pil(image)

        if not return_dict:
            return (image, latent_timestep.item())

        return ImagePipelineOutput(images=image)