modules/flow_controlnet.py

# Copyright 2023 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.
# 
#  Changes were made to this source code by Yuwei Guo.
from dataclasses import dataclass
from typing import Any, Dict, List, Optional, Tuple, Union

import torch
from diffusers import ModelMixin
from diffusers.configuration_utils import ConfigMixin, register_to_config
from diffusers.models.attention_processor import AttentionProcessor
from diffusers.models.embeddings import TimestepEmbedding, Timesteps
from diffusers.models.unet_2d_condition import UNet2DConditionModel
from diffusers.loaders import UNet2DConditionLoadersMixin, PeftAdapterMixin
from diffusers.utils import BaseOutput, logging
from einops import rearrange, repeat
from torch import nn
from torch.nn import functional as F

from .resnet import InflatedConv3d
from .unet_blocks import (
    CrossAttnDownBlock3D,
    DownBlock3D,
    UNetMidBlock3DCrossAttn,
    get_down_block,
)


logger = logging.get_logger(__name__)  # pylint: disable=invalid-name


@dataclass
class FlowControlNetOutput(BaseOutput):
    down_block_res_samples: Tuple[torch.Tensor]
    mid_block_res_sample: torch.Tensor


class FlowControlNetConditioningEmbedding(nn.Module):
    def __init__(
        self,
        conditioning_embedding_channels: int,
        conditioning_channels: int = 3,
        block_out_channels: Tuple[int] = (16, 32, 96, 256),
    ):
        super().__init__()

        self.conv_in = InflatedConv3d(conditioning_channels, block_out_channels[0], kernel_size=3, padding=1)

        self.blocks = nn.ModuleList([])

        for i in range(len(block_out_channels) - 1):
            channel_in = block_out_channels[i]
            channel_out = block_out_channels[i + 1]
            self.blocks.append(InflatedConv3d(channel_in, channel_in, kernel_size=3, padding=1))
            self.blocks.append(InflatedConv3d(channel_in, channel_out, kernel_size=3, padding=1, stride=2))

        self.conv_out = zero_module(
            InflatedConv3d(block_out_channels[-1], conditioning_embedding_channels, kernel_size=3, padding=1)
        )

    def forward(self, conditioning):
        embedding = self.conv_in(conditioning)
        embedding = F.silu(embedding)

        for block in self.blocks:
            embedding = block(embedding)
            embedding = F.silu(embedding)

        embedding = self.conv_out(embedding)

        return embedding


class FlowControlNetModel(ModelMixin, ConfigMixin, UNet2DConditionLoadersMixin, PeftAdapterMixin):
    _supports_gradient_checkpointing = True

    @register_to_config
    def __init__(
        self,
        in_channels: int = 4,
        conditioning_channels: int = 3,
        flip_sin_to_cos: bool = True,
        freq_shift: int = 0,
        down_block_types: Tuple[str] = (
            "CrossAttnDownBlock2D",
            "CrossAttnDownBlock2D",
            "CrossAttnDownBlock2D",
            "DownBlock2D",
        ),
        only_cross_attention: Union[bool, Tuple[bool]] = False,
        block_out_channels: Tuple[int] = (320, 640, 1280, 1280),
        layers_per_block: int = 2,
        downsample_padding: int = 1,
        mid_block_scale_factor: float = 1,
        act_fn: str = "silu",
        norm_num_groups: Optional[int] = 32,
        norm_eps: float = 1e-5,
        cross_attention_dim: int = 1280,
        attention_head_dim: Union[int, Tuple[int]] = 8,
        num_attention_heads: Optional[Union[int, Tuple[int]]] = None,
        use_linear_projection: bool = False,
        class_embed_type: Optional[str] = None,
        num_class_embeds: Optional[int] = None,
        upcast_attention: bool = False,
        resnet_time_scale_shift: str = "default",
        projection_class_embeddings_input_dim: Optional[int] = None,
        controlnet_conditioning_channel_order: str = "rgb",
        conditioning_embedding_out_channels: Optional[Tuple[int]] = (16, 32, 96, 256),
        global_pool_conditions: bool = False,

        use_motion_module         = True,
        motion_module_resolutions = ( 1,2,4,8 ),
        motion_module_mid_block   = False,
        motion_module_type        = "Vanilla",
        motion_module_kwargs      = {
            "num_attention_heads": 8,
            "num_transformer_block": 1,
            "attention_block_types": ["Temporal_Self"],
            "temporal_position_encoding": True,
            "temporal_position_encoding_max_len": 32,
            "temporal_attention_dim_div": 1,
            "causal_temporal_attention": False,
        },

        concate_conditioning_mask: bool = True,
        use_simplified_condition_embedding:  bool = False,

        set_noisy_sample_input_to_zero: bool = False,
    ):
        super().__init__()

        # If `num_attention_heads` is not defined (which is the case for most models)
        # it will default to `attention_head_dim`. This looks weird upon first reading it and it is.
        # The reason for this behavior is to correct for incorrectly named variables that were introduced
        # when this library was created. The incorrect naming was only discovered much later in https://github.com/huggingface/diffusers/issues/2011#issuecomment-1547958131
        # Changing `attention_head_dim` to `num_attention_heads` for 40,000+ configurations is too backwards breaking
        # which is why we correct for the naming here.
        num_attention_heads = num_attention_heads or attention_head_dim

        # Check inputs
        if len(block_out_channels) != len(down_block_types):
            raise ValueError(
                f"Must provide the same number of `block_out_channels` as `down_block_types`. `block_out_channels`: {block_out_channels}. `down_block_types`: {down_block_types}."
            )

        if not isinstance(only_cross_attention, bool) and len(only_cross_attention) != len(down_block_types):
            raise ValueError(
                f"Must provide the same number of `only_cross_attention` as `down_block_types`. `only_cross_attention`: {only_cross_attention}. `down_block_types`: {down_block_types}."
            )

        if not isinstance(num_attention_heads, int) and len(num_attention_heads) != len(down_block_types):
            raise ValueError(
                f"Must provide the same number of `num_attention_heads` as `down_block_types`. `num_attention_heads`: {num_attention_heads}. `down_block_types`: {down_block_types}."
            )

        # input
        self.set_noisy_sample_input_to_zero  = set_noisy_sample_input_to_zero

        conv_in_kernel = 3
        conv_in_padding = (conv_in_kernel - 1) // 2
        self.conv_in = InflatedConv3d(
            in_channels, block_out_channels[0], kernel_size=conv_in_kernel, padding=conv_in_padding
        )
        conditioning_channels = conditioning_channels * 8 * 8
        if concate_conditioning_mask:
            conditioning_channels = conditioning_channels + 1
        self.concate_conditioning_mask = concate_conditioning_mask

        # control net conditioning embedding
        if use_simplified_condition_embedding:
            self.controlnet_cond_embedding = zero_module(
                InflatedConv3d(conditioning_channels, block_out_channels[0], kernel_size=conv_in_kernel, padding=conv_in_padding)
            )
        else:
            self.controlnet_cond_embedding = FlowControlNetConditioningEmbedding(
                conditioning_embedding_channels=block_out_channels[0],
                block_out_channels=conditioning_embedding_out_channels,
                conditioning_channels=conditioning_channels,
            )
        self.use_simplified_condition_embedding = use_simplified_condition_embedding

        # time
        time_embed_dim = block_out_channels[0] * 4

        self.time_proj = Timesteps(block_out_channels[0], flip_sin_to_cos, freq_shift)
        timestep_input_dim = block_out_channels[0]

        self.time_embedding = TimestepEmbedding(
            timestep_input_dim,
            time_embed_dim,
            act_fn=act_fn,
        )

        # class embedding
        if class_embed_type is None and num_class_embeds is not None:
            self.class_embedding = nn.Embedding(num_class_embeds, time_embed_dim)
        elif class_embed_type == "timestep":
            self.class_embedding = TimestepEmbedding(timestep_input_dim, time_embed_dim)
        elif class_embed_type == "identity":
            self.class_embedding = nn.Identity(time_embed_dim, time_embed_dim)
        elif class_embed_type == "projection":
            if projection_class_embeddings_input_dim is None:
                raise ValueError(
                    "`class_embed_type`: 'projection' requires `projection_class_embeddings_input_dim` be set"
                )
            # The projection `class_embed_type` is the same as the timestep `class_embed_type` except
            # 1. the `class_labels` inputs are not first converted to sinusoidal embeddings
            # 2. it projects from an arbitrary input dimension.
            #
            # Note that `TimestepEmbedding` is quite general, being mainly linear layers and activations.
            # When used for embedding actual timesteps, the timesteps are first converted to sinusoidal embeddings.
            # As a result, `TimestepEmbedding` can be passed arbitrary vectors.
            self.class_embedding = TimestepEmbedding(projection_class_embeddings_input_dim, time_embed_dim)
        else:
            self.class_embedding = None


        self.down_blocks = nn.ModuleList([])
        self.controlnet_down_blocks = nn.ModuleList([])

        if isinstance(only_cross_attention, bool):
            only_cross_attention = [only_cross_attention] * len(down_block_types)

        if isinstance(attention_head_dim, int):
            attention_head_dim = (attention_head_dim,) * len(down_block_types)

        if isinstance(num_attention_heads, int):
            num_attention_heads = (num_attention_heads,) * len(down_block_types)

        # down
        output_channel = block_out_channels[0]

        controlnet_block = InflatedConv3d(output_channel, output_channel, kernel_size=1)
        controlnet_block = zero_module(controlnet_block)
        self.controlnet_down_blocks.append(controlnet_block)

        for i, down_block_type in enumerate(down_block_types):
            res = 2 ** i
            input_channel = output_channel
            output_channel = block_out_channels[i]
            is_final_block = i == len(block_out_channels) - 1

            down_block = get_down_block(
                down_block_type,
                num_layers=layers_per_block,
                in_channels=input_channel,
                out_channels=output_channel,
                temb_channels=time_embed_dim,
                add_downsample=not is_final_block,
                resnet_eps=norm_eps,
                resnet_act_fn=act_fn,
                resnet_groups=norm_num_groups,
                cross_attention_dim=cross_attention_dim,
                attn_num_head_channels=attention_head_dim[i] if attention_head_dim[i] is not None else output_channel,
                downsample_padding=downsample_padding,
                use_linear_projection=use_linear_projection,
                only_cross_attention=only_cross_attention[i],
                upcast_attention=upcast_attention,
                resnet_time_scale_shift=resnet_time_scale_shift,

                use_inflated_groupnorm=True,

                use_motion_module=use_motion_module and (res in motion_module_resolutions),
                motion_module_type=motion_module_type,
                motion_module_kwargs=motion_module_kwargs,
            )
            self.down_blocks.append(down_block)

            for _ in range(layers_per_block):
                controlnet_block = InflatedConv3d(output_channel, output_channel, kernel_size=1)
                controlnet_block = zero_module(controlnet_block)
                self.controlnet_down_blocks.append(controlnet_block)

            if not is_final_block:
                controlnet_block = InflatedConv3d(output_channel, output_channel, kernel_size=1)
                controlnet_block = zero_module(controlnet_block)
                self.controlnet_down_blocks.append(controlnet_block)

        # mid
        mid_block_channel = block_out_channels[-1]

        controlnet_block = InflatedConv3d(mid_block_channel, mid_block_channel, kernel_size=1)
        controlnet_block = zero_module(controlnet_block)
        self.controlnet_mid_block = controlnet_block

        self.mid_block = UNetMidBlock3DCrossAttn(
            in_channels=mid_block_channel,
            temb_channels=time_embed_dim,
            resnet_eps=norm_eps,
            resnet_act_fn=act_fn,
            output_scale_factor=mid_block_scale_factor,
            resnet_time_scale_shift=resnet_time_scale_shift,
            cross_attention_dim=cross_attention_dim,
            attn_num_head_channels=num_attention_heads[-1],
            resnet_groups=norm_num_groups,
            use_linear_projection=use_linear_projection,
            upcast_attention=upcast_attention,

            use_inflated_groupnorm=True,
            use_motion_module=use_motion_module and motion_module_mid_block,
            motion_module_type=motion_module_type,
            motion_module_kwargs=motion_module_kwargs,
        )

    @classmethod
    def from_unet(
        cls,
        unet: UNet2DConditionModel,
        controlnet_conditioning_channel_order: str = "rgb",
        conditioning_embedding_out_channels: Optional[Tuple[int]] = (16, 32, 96, 256),
        load_weights_from_unet: bool = True,

        controlnet_additional_kwargs: dict = {},
    ):
        controlnet = cls(
            in_channels=unet.config.in_channels,
            flip_sin_to_cos=unet.config.flip_sin_to_cos,
            freq_shift=unet.config.freq_shift,
            down_block_types=unet.config.down_block_types,
            only_cross_attention=unet.config.only_cross_attention,
            block_out_channels=unet.config.block_out_channels,
            layers_per_block=unet.config.layers_per_block,
            downsample_padding=unet.config.downsample_padding,
            mid_block_scale_factor=unet.config.mid_block_scale_factor,
            act_fn=unet.config.act_fn,
            norm_num_groups=unet.config.norm_num_groups,
            norm_eps=unet.config.norm_eps,
            cross_attention_dim=unet.config.cross_attention_dim,
            attention_head_dim=unet.config.attention_head_dim,
            num_attention_heads=unet.config.num_attention_heads,
            use_linear_projection=unet.config.use_linear_projection,
            class_embed_type=unet.config.class_embed_type,
            num_class_embeds=unet.config.num_class_embeds,
            upcast_attention=unet.config.upcast_attention,
            resnet_time_scale_shift=unet.config.resnet_time_scale_shift,
            projection_class_embeddings_input_dim=unet.config.projection_class_embeddings_input_dim,
            controlnet_conditioning_channel_order=controlnet_conditioning_channel_order,
            conditioning_embedding_out_channels=conditioning_embedding_out_channels,

            **controlnet_additional_kwargs,
        )
        controlnet.unshuffle = nn.PixelUnshuffle(8)

        if load_weights_from_unet:
            m, u = controlnet.conv_in.load_state_dict(cls.image_layer_filter(unet.conv_in.state_dict()), strict=False)
            assert len(u) == 0
            m, u = controlnet.time_proj.load_state_dict(cls.image_layer_filter(unet.time_proj.state_dict()), strict=False)
            assert len(u) == 0
            m, u = controlnet.time_embedding.load_state_dict(cls.image_layer_filter(unet.time_embedding.state_dict()), strict=False)
            assert len(u) == 0

            if controlnet.class_embedding:
                m, u = controlnet.class_embedding.load_state_dict(cls.image_layer_filter(unet.class_embedding.state_dict()), strict=False)
                assert len(u) == 0
            m, u = controlnet.down_blocks.load_state_dict(cls.image_layer_filter(unet.down_blocks.state_dict()), strict=False)
            assert len(u) == 0
            m, u = controlnet.mid_block.load_state_dict(cls.image_layer_filter(unet.mid_block.state_dict()), strict=False)
            assert len(u) == 0

        
        return controlnet

    @staticmethod
    def image_layer_filter(state_dict):
        new_state_dict = {}
        for name, param in state_dict.items():
            if "motion_modules." in name or "lora" in name: continue
            new_state_dict[name] = param
        return new_state_dict

    # Copied from diffusers.models.unet_2d_condition.UNet2DConditionModel.set_attention_slice
    def set_attention_slice(self, slice_size):
        r"""
        Enable sliced attention computation.

        When this option is enabled, the attention module splits the input tensor in slices to compute attention in
        several steps. This is useful for saving some memory in exchange for a small decrease in speed.

        Args:
            slice_size (`str` or `int` or `list(int)`, *optional*, defaults to `"auto"`):
                When `"auto"`, input to the attention heads is halved, so attention is computed in two steps. If
                `"max"`, maximum amount of memory is saved by running only one slice at a time. If a number is
                provided, uses as many slices as `attention_head_dim // slice_size`. In this case, `attention_head_dim`
                must be a multiple of `slice_size`.
        """
        sliceable_head_dims = []

        def fn_recursive_retrieve_sliceable_dims(module: torch.nn.Module):
            if hasattr(module, "set_attention_slice"):
                sliceable_head_dims.append(module.sliceable_head_dim)

            for child in module.children():
                fn_recursive_retrieve_sliceable_dims(child)

        # retrieve number of attention layers
        for module in self.children():
            fn_recursive_retrieve_sliceable_dims(module)

        num_sliceable_layers = len(sliceable_head_dims)

        if slice_size == "auto":
            # half the attention head size is usually a good trade-off between
            # speed and memory
            slice_size = [dim // 2 for dim in sliceable_head_dims]
        elif slice_size == "max":
            # make smallest slice possible
            slice_size = num_sliceable_layers * [1]

        slice_size = num_sliceable_layers * [slice_size] if not isinstance(slice_size, list) else slice_size

        if len(slice_size) != len(sliceable_head_dims):
            raise ValueError(
                f"You have provided {len(slice_size)}, but {self.config} has {len(sliceable_head_dims)} different"
                f" attention layers. Make sure to match `len(slice_size)` to be {len(sliceable_head_dims)}."
            )

        for i in range(len(slice_size)):
            size = slice_size[i]
            dim = sliceable_head_dims[i]
            if size is not None and size > dim:
                raise ValueError(f"size {size} has to be smaller or equal to {dim}.")

        # Recursively walk through all the children.
        # Any children which exposes the set_attention_slice method
        # gets the message
        def fn_recursive_set_attention_slice(module: torch.nn.Module, slice_size: List[int]):
            if hasattr(module, "set_attention_slice"):
                module.set_attention_slice(slice_size.pop())

            for child in module.children():
                fn_recursive_set_attention_slice(child, slice_size)

        reversed_slice_size = list(reversed(slice_size))
        for module in self.children():
            fn_recursive_set_attention_slice(module, reversed_slice_size)

    def _set_gradient_checkpointing(self, module, value=False):
        if isinstance(module, (CrossAttnDownBlock2D, DownBlock2D)):
            module.gradient_checkpointing = value

    def forward(
        self,
        sample: torch.FloatTensor,
        timestep: Union[torch.Tensor, float, int],
        encoder_hidden_states: torch.Tensor,

        controlnet_cond: torch.FloatTensor,
        conditioning_mask: Optional[torch.FloatTensor] = None,

        conditioning_scale: float = 1.0,
        class_labels: Optional[torch.Tensor] = None,
        attention_mask: Optional[torch.Tensor] = None,
        cross_attention_kwargs: Optional[Dict[str, Any]] = None,
        guess_mode: bool = False,
        return_dict: bool = True,
    ) -> Union[FlowControlNetOutput, Tuple]:
        # set input noise to zero
        if self.set_noisy_sample_input_to_zero:
            sample = torch.zeros_like(sample).to(sample.device)
            
        # prepare attention_mask
        if attention_mask is not None:
            attention_mask = (1 - attention_mask.to(sample.dtype)) * -10000.0
            attention_mask = attention_mask.unsqueeze(1)

        # 1. time
        timesteps = timestep
        if not torch.is_tensor(timesteps):
            # TODO: this requires sync between CPU and GPU. So try to pass timesteps as tensors if you can
            # This would be a good case for the `match` statement (Python 3.10+)
            is_mps = sample.device.type == "mps"
            if isinstance(timestep, float):
                dtype = torch.float32 if is_mps else torch.float64
            else:
                dtype = torch.int32 if is_mps else torch.int64
            timesteps = torch.tensor([timesteps], dtype=dtype, device=sample.device)
        elif len(timesteps.shape) == 0:
            timesteps = timesteps[None].to(sample.device)

        timesteps             = timesteps.repeat(sample.shape[0] // timesteps.shape[0])
        encoder_hidden_states = encoder_hidden_states.repeat(sample.shape[0] // encoder_hidden_states.shape[0], 1, 1)

        # broadcast to batch dimension in a way that's compatible with ONNX/Core ML
        timesteps = timesteps.expand(sample.shape[0])

        t_emb = self.time_proj(timesteps)

        # timesteps does not contain any weights and will always return f32 tensors
        # but time_embedding might actually be running in fp16. so we need to cast here.
        # there might be better ways to encapsulate this.
        t_emb = t_emb.to(dtype=self.dtype)
        emb = self.time_embedding(t_emb)

        if self.class_embedding is not None:
            if class_labels is None:
                raise ValueError("class_labels should be provided when num_class_embeds > 0")

            if self.config.class_embed_type == "timestep":
                class_labels = self.time_proj(class_labels)

            class_emb = self.class_embedding(class_labels).to(dtype=self.dtype)
            emb = emb + class_emb

        # 2. pre-process
        sample = self.conv_in(sample)
        

        if self.concate_conditioning_mask:
            controlnet_cond = torch.cat([controlnet_cond, conditioning_mask], dim=1)
        controlnet_cond = self.unshuffle(controlnet_cond.permute(0,2,1,3,4))
        controlnet_cond = controlnet_cond.contiguous().permute(0,2,1,3,4)
        controlnet_cond = self.controlnet_cond_embedding(controlnet_cond)
        
        sample = sample + controlnet_cond

        # 3. down
        down_block_res_samples = (sample,)
        for downsample_block in self.down_blocks:
            if hasattr(downsample_block, "has_cross_attention") and downsample_block.has_cross_attention:
                sample, res_samples = downsample_block(
                    hidden_states=sample,
                    temb=emb,
                    encoder_hidden_states=encoder_hidden_states,
                    attention_mask=attention_mask,
                    # cross_attention_kwargs=cross_attention_kwargs,
                )
            else: sample, res_samples = downsample_block(hidden_states=sample, temb=emb)

            down_block_res_samples += res_samples

        # 4. mid
        if self.mid_block is not None:
            sample = self.mid_block(
                sample,
                emb,
                encoder_hidden_states=encoder_hidden_states,
                attention_mask=attention_mask,
                # cross_attention_kwargs=cross_attention_kwargs,
            )

        # 5. controlnet blocks
        controlnet_down_block_res_samples = ()

        for down_block_res_sample, controlnet_block in zip(down_block_res_samples, self.controlnet_down_blocks):
            down_block_res_sample = controlnet_block(down_block_res_sample)
            controlnet_down_block_res_samples = controlnet_down_block_res_samples + (down_block_res_sample,)

        down_block_res_samples = controlnet_down_block_res_samples

        mid_block_res_sample = self.controlnet_mid_block(sample)

        # 6. scaling
        if guess_mode and not self.config.global_pool_conditions:
            scales = torch.logspace(-1, 0, len(down_block_res_samples) + 1, device=sample.device)  # 0.1 to 1.0

            scales = scales * conditioning_scale
            down_block_res_samples = [sample * scale for sample, scale in zip(down_block_res_samples, scales)]
            mid_block_res_sample = mid_block_res_sample * scales[-1]  # last one
        else:
            down_block_res_samples = [sample * conditioning_scale for sample in down_block_res_samples]
            mid_block_res_sample = mid_block_res_sample * conditioning_scale

        if self.config.global_pool_conditions:
            down_block_res_samples = [
                torch.mean(sample, dim=(2, 3), keepdim=True) for sample in down_block_res_samples
            ]
            mid_block_res_sample = torch.mean(mid_block_res_sample, dim=(2, 3), keepdim=True)

        if not return_dict:
            return (down_block_res_samples, mid_block_res_sample)

        return FlowControlNetOutput(
            down_block_res_samples=down_block_res_samples, mid_block_res_sample=mid_block_res_sample
        )


def zero_module(module):
    for p in module.parameters():
        nn.init.zeros_(p)
    return module