pipeline_ddim.py

# 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 Optional, Tuple, Union

import torch

from ...pipeline_utils import DiffusionPipeline, ImagePipelineOutput
from ...utils import deprecate


class DDIMPipeline(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__()
        self.register_modules(unet=unet, scheduler=scheduler)

    @torch.no_grad()
    def __call__(
        self,
        batch_size: int = 1,
        generator: Optional[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:
            batch_size (`int`, *optional*, defaults to 1):
                The number of images to generate.
            generator (`torch.Generator`, *optional*):
                A [torch generator](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 [`~pipeline_utils.ImagePipelineOutput`] instead of a plain tuple.

        Returns:
            [`~pipeline_utils.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.
        """

        if generator is not None and generator.device.type != self.device.type and self.device.type != "mps":
            message = (
                f"The `generator` device is `{generator.device}` and does not match the pipeline "
                f"device `{self.device}`, so the `generator` will be ignored. "
                f'Please use `generator=torch.Generator(device="{self.device}")` instead.'
            )
            deprecate(
                "generator.device == 'cpu'",
                "0.12.0",
                message,
            )
            generator = None

        # Sample gaussian noise to begin loop
        if isinstance(self.unet.sample_size, int):
            image_shape = (batch_size, self.unet.in_channels, self.unet.sample_size, self.unet.sample_size)
        else:
            image_shape = (batch_size, self.unet.in_channels, *self.unet.sample_size)

        if self.device.type == "mps":
            # randn does not work reproducibly on mps
            image = torch.randn(image_shape, generator=generator, dtype=self.unet.dtype)
            image = image.to(self.device)
        else:
            image = torch.randn(image_shape, generator=generator, device=self.device, dtype=self.unet.dtype)

        # set step values
        self.scheduler.set_timesteps(num_inference_steps)

        for t in self.progress_bar(self.scheduler.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,)

        return ImagePipelineOutput(images=image)