gqvr/model/controlnet.py

import einops
import torch
import torch as th
import torch.nn as nn

from .util import (
    conv_nd,
    linear,
    zero_module,
    timestep_embedding,
)

from .attention import SpatialTransformer, liem
from .openaimodel import UNetModel, TimestepEmbedSequential, ResBlock, Downsample, AttentionBlock, Upsample

def exists(x):
    return x is not None


class ZeroSRCConnector(nn.Module):
    def __init__(self, in_channels, feat_channels):
        super().__init__()
        self.norm = nn.GroupNorm(32, feat_channels)
        self.scale = nn.Sequential(
            nn.Conv2d(in_channels, feat_channels, 3, padding=1),
            nn.ReLU(),
            nn.Conv2d(feat_channels, feat_channels, 3, padding=1),
        )
        self.shift = nn.Sequential(
            nn.Conv2d(in_channels, feat_channels, 3, padding=1),
            nn.ReLU(),
            nn.Conv2d(feat_channels, feat_channels, 3, padding=1),
        )

    def forward(self, x, feat, control_scale=1.0):
        x = self.norm(x)
        scale = self.scale(feat) * control_scale
        shift = self.shift(feat) * control_scale
        return x * (1 + scale) + shift
    

class ControlledUnetModel(UNetModel):
    def __init__(self, *args, use_liem=True, liem_kernel_size=7, **kwargs):
        super().__init__(*args, **kwargs)
        self.use_liem = use_liem
        self.liem_modules = nn.ModuleList()
        self.liem_indices = []  # Keep track of which decoder blocks get LIEM

        for i, block in enumerate(self.output_blocks):
            if any(isinstance(layer, SpatialTransformer) for layer in block[:2]):
                self.liem_modules.append(liem(liem_kernel_size))
                self.liem_indices.append(i)

        self.zeroSRC_connector = nn.ModuleList()
        for module in self.output_blocks:
            if len(module) == 3:
                ch = module[0].out_channels
                self.zeroSRC_connector.append(ZeroSRCConnector(ch, ch))  # before upsample
            ch = module[0].out_channels
            self.zeroSRC_connector.append(ZeroSRCConnector(ch, ch))    # after skip


    def forward(self, x, timesteps=None, context=None, control=None, only_mid_control=False, **kwargs):
        hs = []
        with torch.no_grad():
            t_emb = timestep_embedding(timesteps, self.model_channels, repeat_only=False)
            emb = self.time_embed(t_emb)
            h = x.type(self.dtype)
            for module in self.input_blocks:
                h = module(h, emb, context)
                hs.append(h)

            adapter_idx = len(self.zeroSRC_connector) - 1
            control_idx = len(control) - 1 if control else None
            h = self.middle_block(h, emb, context)

        if control is not None:
            combined = control[control_idx] + hs.pop()
            h = self.zeroSRC_connector[adapter_idx](h, combined, control_scale=1.)
            adapter_idx -= 1
            control_idx -= 1

        for i, module in enumerate(self.output_blocks):
            skip = hs.pop()
            # h = self.project_modules[adapter_idx](control[control_idx], skip, h)
            # adapter_idx -= 1
            if len(module) == 3:
                assert isinstance(module[2], Upsample)
                # Run through res blk & transformer blk
                for layer in module[:2]:
                    if isinstance(layer, ResBlock):
                        h = layer(h, emb)
                    elif isinstance(layer, SpatialTransformer):
                        if self.use_liem and i in self.liem_indices:
                            liem_idx = self.liem_indices.index(i)
                            h = self.liem_modules[liem_idx](h)
                        h = layer(h, context)
                    else:
                        h = layer(h)

                if control is not None:
                    combined = control[control_idx] + skip
                    h = self.zeroSRC_connector[adapter_idx](h, combined, control_scale=1.)
                    adapter_idx -= 1
                    control_idx -= 1
                h = module[2](h)  # Upsample 
                    
                # Second control injection after skip connection
                if only_mid_control or control is None:
                    # h = torch.cat([h, hs.pop()], dim=1)
                    h = self.zeroSRC_connector[adapter_idx](h, skip, control_scale=1.)
                else:
                    combined = control[control_idx] + skip
                    h = self.zeroSRC_connector[adapter_idx](h, combined, control_scale=1.)
                    control_idx -= 1
                adapter_idx -= 1

            else:
                if only_mid_control or control is None:
                    h = self.zeroSRC_connector[adapter_idx](h, skip, control_scale=1.)
                else:
                    combined = control[control_idx] + skip
                    h = self.zeroSRC_connector[adapter_idx](h, combined, control_scale=1.)
                    control_idx -= 1
                adapter_idx -= 1
                
                h = module(h, emb, context)

        h = h.type(x.dtype)
        return self.out(h)


class ControlNet(nn.Module):
    def __init__(
            self,
            image_size,
            in_channels,
            model_channels,
            hint_channels,
            num_res_blocks,
            attention_resolutions,
            dropout=0,
            channel_mult=(1, 2, 4, 8),
            conv_resample=True,
            dims=2,
            use_checkpoint=False,
            use_fp16=False,
            num_heads=-1,
            num_head_channels=-1,
            num_heads_upsample=-1,
            use_scale_shift_norm=False,
            resblock_updown=False,
            use_new_attention_order=False,
            use_spatial_transformer=False,  # custom transformer support
            transformer_depth=1,  # custom transformer support
            context_dim=None,  # custom transformer support
            n_embed=None,  # custom support for prediction of discrete ids into codebook of first stage vq model
            legacy=True,
            disable_self_attentions=None,
            num_attention_blocks=None,
            disable_middle_self_attn=False,
            use_linear_in_transformer=False,
    ):
        super().__init__()
        if use_spatial_transformer:
            assert context_dim is not None, 'You forgot to include the dimension of your cross-attention conditioning...'

        if context_dim is not None:
            assert use_spatial_transformer, 'You forgot to use the spatial transformer for your cross-attention conditioning...'
            from omegaconf.listconfig import ListConfig
            if type(context_dim) == ListConfig:
                context_dim = list(context_dim)

        if num_heads_upsample == -1:
            num_heads_upsample = num_heads

        if num_heads == -1:
            assert num_head_channels != -1, 'Either num_heads or num_head_channels has to be set'

        if num_head_channels == -1:
            assert num_heads != -1, 'Either num_heads or num_head_channels has to be set'

        self.dims = dims
        self.image_size = image_size
        self.in_channels = in_channels
        self.model_channels = model_channels
        if isinstance(num_res_blocks, int):
            self.num_res_blocks = len(channel_mult) * [num_res_blocks]
        else:
            if len(num_res_blocks) != len(channel_mult):
                raise ValueError("provide num_res_blocks either as an int (globally constant) or "
                                 "as a list/tuple (per-level) with the same length as channel_mult")
            self.num_res_blocks = num_res_blocks
        if disable_self_attentions is not None:
            # should be a list of booleans, indicating whether to disable self-attention in TransformerBlocks or not
            assert len(disable_self_attentions) == len(channel_mult)
        if num_attention_blocks is not None:
            assert len(num_attention_blocks) == len(self.num_res_blocks)
            assert all(map(lambda i: self.num_res_blocks[i] >= num_attention_blocks[i], range(len(num_attention_blocks))))
            print(f"Constructor of UNetModel received num_attention_blocks={num_attention_blocks}. "
                  f"This option has LESS priority than attention_resolutions {attention_resolutions}, "
                  f"i.e., in cases where num_attention_blocks[i] > 0 but 2**i not in attention_resolutions, "
                  f"attention will still not be set.")

        self.attention_resolutions = attention_resolutions
        self.dropout = dropout
        self.channel_mult = channel_mult
        self.conv_resample = conv_resample
        self.use_checkpoint = use_checkpoint
        self.dtype = th.float16 if use_fp16 else th.float32
        self.num_heads = num_heads
        self.num_head_channels = num_head_channels
        self.num_heads_upsample = num_heads_upsample
        self.predict_codebook_ids = n_embed is not None

        time_embed_dim = model_channels * 4
        self.time_embed = nn.Sequential(
            linear(model_channels, time_embed_dim),
            nn.SiLU(),
            linear(time_embed_dim, time_embed_dim),
        )

        self.input_blocks = nn.ModuleList(
            [
                TimestepEmbedSequential(
                    conv_nd(dims, in_channels, model_channels, 3, padding=1)
                )
            ]
        )
        self.zero_convs = nn.ModuleList([self.make_zero_conv(model_channels)])

        self.input_hint_block = TimestepEmbedSequential(
            conv_nd(dims, hint_channels, 16, 3, padding=1),
            nn.SiLU(),
            conv_nd(dims, 16, 16, 3, padding=1),
            nn.SiLU(),
            conv_nd(dims, 16, 32, 3, padding=1, stride=2),
            nn.SiLU(),
            conv_nd(dims, 32, 32, 3, padding=1),
            nn.SiLU(),
            conv_nd(dims, 32, 96, 3, padding=1, stride=2),
            nn.SiLU(),
            conv_nd(dims, 96, 96, 3, padding=1),
            nn.SiLU(),
            conv_nd(dims, 96, 256, 3, padding=1, stride=2),
            nn.SiLU(),
            zero_module(conv_nd(dims, 256, model_channels, 3, padding=1))
        )

        self._feature_size = model_channels
        input_block_chans = [model_channels]
        ch = model_channels
        ds = 1
        for level, mult in enumerate(channel_mult):
            for nr in range(self.num_res_blocks[level]):
                layers = [
                    ResBlock(
                        ch,
                        time_embed_dim,
                        dropout,
                        out_channels=mult * model_channels,
                        dims=dims,
                        use_checkpoint=use_checkpoint,
                        use_scale_shift_norm=use_scale_shift_norm,
                    )
                ]
                ch = mult * model_channels
                if ds in attention_resolutions:
                    if num_head_channels == -1:
                        dim_head = ch // num_heads
                    else:
                        num_heads = ch // num_head_channels
                        dim_head = num_head_channels
                    if legacy:
                        # num_heads = 1
                        dim_head = ch // num_heads if use_spatial_transformer else num_head_channels
                    if exists(disable_self_attentions):
                        disabled_sa = disable_self_attentions[level]
                    else:
                        disabled_sa = False

                    if not exists(num_attention_blocks) or nr < num_attention_blocks[level]:
                        layers.append(
                            AttentionBlock(
                                ch,
                                use_checkpoint=use_checkpoint,
                                num_heads=num_heads,
                                num_head_channels=dim_head,
                                use_new_attention_order=use_new_attention_order,
                            ) if not use_spatial_transformer else SpatialTransformer(
                                ch, num_heads, dim_head, depth=transformer_depth, context_dim=context_dim,
                                disable_self_attn=disabled_sa, use_linear=use_linear_in_transformer,
                                use_checkpoint=use_checkpoint
                            )
                        )
                self.input_blocks.append(TimestepEmbedSequential(*layers))
                self.zero_convs.append(self.make_zero_conv(ch))
                self._feature_size += ch
                input_block_chans.append(ch)
            if level != len(channel_mult) - 1:
                out_ch = ch
                self.input_blocks.append(
                    TimestepEmbedSequential(
                        ResBlock(
                            ch,
                            time_embed_dim,
                            dropout,
                            out_channels=out_ch,
                            dims=dims,
                            use_checkpoint=use_checkpoint,
                            use_scale_shift_norm=use_scale_shift_norm,
                            down=True,
                        )
                        if resblock_updown
                        else Downsample(
                            ch, conv_resample, dims=dims, out_channels=out_ch
                        )
                    )
                )
                ch = out_ch
                input_block_chans.append(ch)
                self.zero_convs.append(self.make_zero_conv(ch))
                ds *= 2
                self._feature_size += ch

        if num_head_channels == -1:
            dim_head = ch // num_heads
        else:
            num_heads = ch // num_head_channels
            dim_head = num_head_channels
        if legacy:
            # num_heads = 1
            dim_head = ch // num_heads if use_spatial_transformer else num_head_channels
        self.middle_block = TimestepEmbedSequential(
            ResBlock(
                ch,
                time_embed_dim,
                dropout,
                dims=dims,
                use_checkpoint=use_checkpoint,
                use_scale_shift_norm=use_scale_shift_norm,
            ),
            AttentionBlock(
                ch,
                use_checkpoint=use_checkpoint,
                num_heads=num_heads,
                num_head_channels=dim_head,
                use_new_attention_order=use_new_attention_order,
            ) if not use_spatial_transformer else SpatialTransformer(  # always uses a self-attn
                ch, num_heads, dim_head, depth=transformer_depth, context_dim=context_dim,
                disable_self_attn=disable_middle_self_attn, use_linear=use_linear_in_transformer,
                use_checkpoint=use_checkpoint
            ),
            ResBlock(
                ch,
                time_embed_dim,
                dropout,
                dims=dims,
                use_checkpoint=use_checkpoint,
                use_scale_shift_norm=use_scale_shift_norm,
            ),
        )
        self.middle_block_out = self.make_zero_conv(ch)
        self._feature_size += ch

    def make_zero_conv(self, channels):
        return TimestepEmbedSequential(zero_module(conv_nd(self.dims, channels, channels, 1, padding=0)))

    def forward(self, x, hint, timesteps, context, **kwargs):
        t_emb = timestep_embedding(timesteps, self.model_channels, repeat_only=False)
        emb = self.time_embed(t_emb)

        guided_hint = self.input_hint_block(hint, emb, context)

        outs = []

        h = x.type(self.dtype)
        for module, zero_conv in zip(self.input_blocks, self.zero_convs):
            if guided_hint is not None:
                h = module(h, emb, context)
                h += guided_hint
                guided_hint = None
            else:
                h = module(h, emb, context)
            outs.append(zero_conv(h, emb, context))

        h = self.middle_block(h, emb, context)
        outs.append(self.middle_block_out(h, emb, context))

        return outs