import torch import torch as th import torch.nn as nn import torch.nn.functional as F from ldm.modules.diffusionmodules.util import ( conv_nd, linear, zero_module, timestep_embedding ) from einops import rearrange from ldm.modules.attention import SpatialTransformer from ldm.modules.diffusionmodules.openaimodel import UNetModel, TimestepEmbedSequential, ResBlock, Downsample, AttentionBlock from ldm.util import exists class StableVITON(UNetModel): def __init__( self, dim_head_denorm=1, *args, **kwargs, ): super().__init__(*args, **kwargs) warp_flow_blks = [] warp_zero_convs = [] self.encode_output_chs = [ 320, 320, 640, 640, 640, 1280, 1280, 1280, 1280 ] self.encode_output_chs2 = [ 320, 320, 320, 320, 640, 640, 640, 1280, 1280 ] for in_ch, cont_ch in zip(self.encode_output_chs, self.encode_output_chs2): dim_head = in_ch // self.num_heads dim_head = dim_head // dim_head_denorm warp_flow_blks.append(SpatialTransformer( in_channels=in_ch, n_heads=self.num_heads, d_head=dim_head, depth=self.transformer_depth, context_dim=cont_ch, use_linear=self.use_linear_in_transformer, use_checkpoint=self.use_checkpoint, )) warp_zero_convs.append(self.make_zero_conv(in_ch)) self.warp_flow_blks = nn.ModuleList(reversed(warp_flow_blks)) self.warp_zero_convs = nn.ModuleList(reversed(warp_zero_convs)) def make_zero_conv(self, channels): return zero_module(conv_nd(2, channels, channels, 1, padding=0)) 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) h = self.middle_block(h, emb, context) if control is not None: hint = control.pop() for module in self.output_blocks[:3]: control.pop() h = torch.cat([h, hs.pop()], dim=1) h = module(h, emb, context) n_warp = len(self.encode_output_chs) for i, (module, warp_blk, warp_zc) in enumerate(zip(self.output_blocks[3:n_warp+3], self.warp_flow_blks, self.warp_zero_convs)): if control is None or (h.shape[-2] == 8 and h.shape[-1] == 6): assert 0, f"shape is wrong : {h.shape}" else: hint = control.pop() h = self.warp(h, hint, warp_blk, warp_zc) h = torch.cat([h, hs.pop()], dim=1) h = module(h, emb, context) for module in self.output_blocks[n_warp+3:]: if control is None: h = torch.cat([h, hs.pop()], dim=1) else: h = torch.cat([h, hs.pop()], dim=1) h = module(h, emb, context) h = h.type(x.dtype) return self.out(h) def warp(self, x, hint, crossattn_layer, zero_conv, mask1=None, mask2=None): hint = rearrange(hint, "b c h w -> b (h w) c").contiguous() output = crossattn_layer(x, hint) output = zero_conv(output) return output + x class NoZeroConvControlNet(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, legacy=True, disable_self_attentions=None, num_attention_blocks=None, disable_middle_self_attn=False, use_linear_in_transformer=False, use_VAEdownsample=False, cond_first_ch=8, ): super().__init__() if use_spatial_transformer: assert context_dim is not None, 'Fool!! You forgot to include the dimension of your cross-attention conditioning...' if context_dim is not None: assert use_spatial_transformer, 'Fool!! 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 um_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 self.use_VAEdownsample = use_VAEdownsample 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.cond_first_block = TimestepEmbedSequential( zero_module(conv_nd(dims, cond_first_ch, 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._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) 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._feature_size += ch def forward(self, x, hint, timesteps, context, only_mid_control=False, **kwargs): t_emb = timestep_embedding(timesteps, self.model_channels, repeat_only=False) emb = self.time_embed(t_emb) if not self.use_VAEdownsample: guided_hint = self.input_hint_block(hint, emb, context) else: guided_hint = self.cond_first_block(hint, emb, context) outs = [] hs = [] h = x.type(self.dtype) for module in self.input_blocks: if guided_hint is not None: h = module(h, emb, context) h += guided_hint hs.append(h) guided_hint = None else: h = module(h, emb, context) hs.append(h) outs.append(h) h = self.middle_block(h, emb, context) outs.append(h) return outs, None