from abc import abstractmethod import math import torch import numpy as np import torch as th import torch.nn as nn import torch.nn.functional as F try: import xformers import xformers.ops XFORMERS_AVAILABLE = True except: XFORMERS_AVAILABLE = False from diffcve.modules.diffusion.utils import checkpoint, conv_nd, linear, avg_pool_nd, zero_module, normalization, timestep_embedding, SpatialTemporalConv from diffcve.modules.attention import SpatialTransformer, ProjectedSpatialTransformer, TemporalAttention from diffcve.modules.spade import SPADE def convert_module_to_f16(x): pass def convert_module_to_f32(x): pass def exists(val): return val is not None class TimestepBlock(nn.Module): @abstractmethod def forward(self, x, emb): pass class StructureConditionedBlock(nn.Module): @abstractmethod def forward(self, x, emb, cond): pass class PriorConditionedBlock(nn.Module): @abstractmethod def forward(self, x, emb, struct_cond, prior_cond): pass class TimestepEmbedSequential(nn.Sequential, TimestepBlock): def forward(self, x, emb, context=None, struct_cond=None, prior_cond=None): for layer in self: if isinstance(layer, TimestepBlock): x = layer(x, emb) elif isinstance(layer, SpatialTransformer) or isinstance(layer, ProjectedSpatialTransformer): assert context is not None x = layer(x, context) elif isinstance(layer, StructureConditionedBlock): assert struct_cond is not None x = layer(x, emb, struct_cond) elif isinstance(layer, PriorConditionedBlock): assert struct_cond is not None assert prior_cond is not None x = layer(x, emb, struct_cond, prior_cond) else: x = layer(x) return x class Upsample(nn.Module): def __init__(self, channels, use_conv, dims=2, out_channels=None, padding=1): super().__init__() self.channels = channels self.out_channels = out_channels or channels self.use_conv = use_conv self.dims = dims if use_conv: self.conv = conv_nd(dims, self.channels, self.out_channels, 3, padding=padding) def forward(self, x): assert x.shape[1] == self.channels if self.dims == 3: x = F.interpolate( x, (x.shape[2], x.shape[3] * 2, x.shape[4] * 2), mode="nearest" ) else: x = F.interpolate(x, scale_factor=2, mode="nearest") if self.use_conv: x = self.conv(x) return x class Downsample(nn.Module): def __init__(self, channels, use_conv, dims=2, out_channels=None,padding=1): super().__init__() self.channels = channels self.out_channels = out_channels or channels self.use_conv = use_conv self.dims = dims stride = 2 if dims != 3 else (1, 2, 2) if use_conv: self.op = conv_nd( dims, self.channels, self.out_channels, 3, stride=stride, padding=padding ) else: assert self.channels == self.out_channels self.op = avg_pool_nd(dims, kernel_size=stride, stride=stride) def forward(self, x): assert x.shape[1] == self.channels return self.op(x) class ResBlock(TimestepBlock): def __init__( self, channels, emb_channels, dropout, out_channels=None, use_conv=False, use_scale_shift_norm=False, dims=2, use_checkpoint=False, up=False, down=False, ): super().__init__() self.channels = channels self.emb_channels = emb_channels self.dropout = dropout self.out_channels = out_channels or channels self.use_conv = use_conv self.use_checkpoint = use_checkpoint self.use_scale_shift_norm = use_scale_shift_norm self.in_layers = nn.Sequential( normalization(channels), nn.SiLU(), conv_nd(dims, channels, self.out_channels, 3, padding=1), ) self.updown = up or down if up: self.h_upd = Upsample(channels, False, dims) self.x_upd = Upsample(channels, False, dims) elif down: self.h_upd = Downsample(channels, False, dims) self.x_upd = Downsample(channels, False, dims) else: self.h_upd = self.x_upd = nn.Identity() self.emb_layers = nn.Sequential( nn.SiLU(), linear( emb_channels, 2 * self.out_channels if use_scale_shift_norm else self.out_channels, ), ) if self.out_channels % 32 == 0: self.out_layers = nn.Sequential( normalization(self.out_channels), nn.SiLU(), nn.Dropout(p=dropout), zero_module( conv_nd(dims, self.out_channels, self.out_channels, 3, padding=1) ), ) else: self.out_layers = nn.Sequential( normalization(self.out_channels, self.out_channels), nn.SiLU(), nn.Dropout(p=dropout), zero_module( conv_nd(dims, self.out_channels, self.out_channels, 3, padding=1) ), ) if self.out_channels == channels: self.skip_connection = nn.Identity() elif use_conv: self.skip_connection = conv_nd( dims, channels, self.out_channels, 3, padding=1 ) else: self.skip_connection = conv_nd(dims, channels, self.out_channels, 1) def forward(self, x, emb): return checkpoint( self._forward, (x, emb), self.parameters(), self.use_checkpoint ) def _forward(self, x, emb): if self.updown: in_rest, in_conv = self.in_layers[:-1], self.in_layers[-1] h = in_rest(x) h = self.h_upd(h) x = self.x_upd(x) h = in_conv(h) else: h = self.in_layers(x) emb_out = self.emb_layers(emb).type(h.dtype) while len(emb_out.shape) < len(h.shape): emb_out = emb_out[..., None] if self.use_scale_shift_norm: out_norm, out_rest = self.out_layers[0], self.out_layers[1:] scale, shift = th.chunk(emb_out, 2, dim=1) h = out_norm(h) * (1 + scale) + shift h = out_rest(h) else: h = h + emb_out h = self.out_layers(h) return self.skip_connection(x) + h class StructureConditionedResBlock(StructureConditionedBlock): def __init__( self, channels, emb_channels, dropout, condition_channels, out_channels=None, use_conv=False, use_scale_shift_norm=False, dims=2, use_checkpoint=False, up=False, down=False, ): super().__init__() self.channels = channels self.emb_channels = emb_channels self.dropout = dropout self.out_channels = out_channels or channels self.use_conv = use_conv self.use_checkpoint = use_checkpoint self.use_scale_shift_norm = use_scale_shift_norm self.in_layers = nn.Sequential( normalization(channels), nn.SiLU(), conv_nd(dims, channels, self.out_channels, 3, padding=1), ) self.updown = up or down if up: self.h_upd = Upsample(channels, False, dims) self.x_upd = Upsample(channels, False, dims) elif down: self.h_upd = Downsample(channels, False, dims) self.x_upd = Downsample(channels, False, dims) else: self.h_upd = self.x_upd = nn.Identity() self.emb_layers = nn.Sequential( nn.SiLU(), linear( emb_channels, 2 * self.out_channels if use_scale_shift_norm else self.out_channels, ), ) self.spade = SPADE(self.out_channels, condition_channels) self.out_layers = nn.Sequential( normalization(self.out_channels), nn.SiLU(), nn.Dropout(p=dropout), zero_module( conv_nd(dims, self.out_channels, self.out_channels, 3, padding=1) ), ) if self.out_channels == channels: self.skip_connection = nn.Identity() elif use_conv: self.skip_connection = conv_nd( dims, channels, self.out_channels, 3, padding=1 ) else: self.skip_connection = conv_nd(dims, channels, self.out_channels, 1) def forward(self, x, emb, s_cond): return checkpoint( self._forward, (x, emb, s_cond), self.parameters(), self.use_checkpoint ) def _forward(self, x, emb, s_cond): if self.updown: in_rest, in_conv = self.in_layers[:-1], self.in_layers[-1] h = in_rest(x) h = self.h_upd(h) x = self.x_upd(x) h = in_conv(h) else: h = self.in_layers(x) emb_out = self.emb_layers(emb).type(h.dtype) while len(emb_out.shape) < len(h.shape): emb_out = emb_out[..., None] if self.use_scale_shift_norm: out_norm, out_rest = self.out_layers[0], self.out_layers[1:] scale, shift = th.chunk(emb_out, 2, dim=1) h = out_norm(h) * (1 + scale) + shift h = out_rest(h) else: h = h + emb_out h = self.out_layers(h) h = self.spade(h, s_cond) return self.skip_connection(x) + h class PriorConditionedResBlock(PriorConditionedBlock): def __init__( self, channels, emb_channels, dropout, condition_channels, out_channels=None, use_conv=False, use_scale_shift_norm=False, dims=2, use_checkpoint=False, up=False, down=False, ): super().__init__() self.channels = channels self.emb_channels = emb_channels self.dropout = dropout self.out_channels = out_channels or channels self.use_conv = use_conv self.use_checkpoint = use_checkpoint self.use_scale_shift_norm = use_scale_shift_norm self.in_layers = nn.Sequential( normalization(channels), nn.SiLU(), conv_nd(dims, channels, self.out_channels, 3, padding=1), ) self.updown = up or down if up: self.h_upd = Upsample(channels, False, dims) self.x_upd = Upsample(channels, False, dims) elif down: self.h_upd = Downsample(channels, False, dims) self.x_upd = Downsample(channels, False, dims) else: self.h_upd = self.x_upd = nn.Identity() self.emb_layers = nn.Sequential( nn.SiLU(), linear( emb_channels, 2 * self.out_channels if use_scale_shift_norm else self.out_channels, ), ) self.s_spade = SPADE(self.out_channels, condition_channels) self.c_spade = SPADE(self.out_channels, condition_channels) self.out_layers = nn.Sequential( normalization(self.out_channels), nn.SiLU(), nn.Dropout(p=dropout), zero_module( conv_nd(dims, self.out_channels, self.out_channels, 3, padding=1) ), ) if self.out_channels == channels: self.skip_connection = nn.Identity() elif use_conv: self.skip_connection = conv_nd( dims, channels, self.out_channels, 3, padding=1 ) else: self.skip_connection = conv_nd(dims, channels, self.out_channels, 1) def forward(self, x, emb, s_cond, p_cond): return checkpoint( self._forward, (x, emb, s_cond, p_cond), self.parameters(), self.use_checkpoint ) def _forward(self, x, emb, s_cond, p_cond): if self.updown: in_rest, in_conv = self.in_layers[:-1], self.in_layers[-1] h = in_rest(x) h = self.h_upd(h) x = self.x_upd(x) h = in_conv(h) else: h = self.in_layers(x) emb_out = self.emb_layers(emb).type(h.dtype) while len(emb_out.shape) < len(h.shape): emb_out = emb_out[..., None] if self.use_scale_shift_norm: out_norm, out_rest = self.out_layers[0], self.out_layers[1:] scale, shift = th.chunk(emb_out, 2, dim=1) h = out_norm(h) * (1 + scale) + shift h = out_rest(h) else: h = h + emb_out h = self.out_layers(h) h = self.c_spade(h, p_cond) h = self.s_spade(h, s_cond) return self.skip_connection(x) + h class AttentionBlock(nn.Module): def __init__( self, channels, num_heads=1, num_head_channels=-1, use_checkpoint=False, use_new_attention_order=False, ): super().__init__() self.channels = channels if num_head_channels == -1: self.num_heads = num_heads else: assert ( channels % num_head_channels == 0 ), f"q,k,v channels {channels} is not divisible by num_head_channels {num_head_channels}" self.num_heads = channels // num_head_channels self.use_checkpoint = use_checkpoint self.norm = normalization(channels) self.qkv = conv_nd(1, channels, channels * 3, 1) if use_new_attention_order: self.attention = QKVAttention(self.num_heads) else: self.attention = QKVAttentionLegacy(self.num_heads) self.proj_out = zero_module(conv_nd(1, channels, channels, 1)) def forward(self, x): return checkpoint(self._forward, (x,), self.parameters(), self.use_checkpoint) def _forward(self, x): b, c, *spatial = x.shape x = x.reshape(b, c, -1) qkv = self.qkv(self.norm(x)) h = self.attention(qkv) h = self.proj_out(h) return (x + h).reshape(b, c, *spatial) def count_flops_attn(model, _x, y): b, c, *spatial = y[0].shape num_spatial = int(np.prod(spatial)) matmul_ops = 2 * b * (num_spatial ** 2) * c model.total_ops += th.DoubleTensor([matmul_ops]) class QKVAttentionLegacy(nn.Module): def __init__(self, n_heads): super().__init__() self.n_heads = n_heads self.attention_op: Optional[Any] = None def forward(self, qkv): bs, width, length = qkv.shape assert width % (3 * self.n_heads) == 0 ch = width // (3 * self.n_heads) q, k, v = qkv.reshape(bs * self.n_heads, ch * 3, length).split(ch, dim=1) scale = 1 / math.sqrt(math.sqrt(ch)) if XFORMERS_AVAILABLE and not torch.onnx.is_in_onnx_export(): q, k, v = map( lambda t:t.permute(0,2,1) .contiguous(), (q, k, v), ) a = xformers.ops.memory_efficient_attention(q, k, v, attn_bias=None, op=self.attention_op) a = ( a.permute(0,2,1) .reshape(bs, -1, length) ) else: weight = th.einsum( "bct,bcs->bts", q * scale, k * scale ) weight = th.softmax(weight.float(), dim=-1).type(weight.dtype) a = th.einsum("bts,bcs->bct", weight, v) a = a.reshape(bs, -1, length) return a @staticmethod def count_flops(model, _x, y): return count_flops_attn(model, _x, y) class QKVAttention(nn.Module): def __init__(self, n_heads): super().__init__() self.n_heads = n_heads self.attention_op: Optional[Any] = None def forward(self, qkv): bs, width, length = qkv.shape assert width % (3 * self.n_heads) == 0 ch = width // (3 * self.n_heads) q, k, v = qkv.chunk(3, dim=1) scale = 1 / math.sqrt(math.sqrt(ch)) if XFORMERS_AVAILABLE and not torch.onnx.is_in_onnx_export(): q, k, v = map( lambda t:t.permute(0,2,1) .contiguous(), (q, k, v), ) a = xformers.ops.memory_efficient_attention(q, k, v, attn_bias=None, op=self.attention_op) a = ( a.permute(0,2,1) .reshape(bs, -1, length) ) else: weight = th.einsum( "bct,bcs->bts", (q * scale).view(bs * self.n_heads, ch, length), (k * scale).view(bs * self.n_heads, ch, length), ) weight = th.softmax(weight.float(), dim=-1).type(weight.dtype) a = th.einsum("bts,bcs->bct", weight, v.reshape(bs * self.n_heads, ch, length)) a = a.reshape(bs, -1, length) return a @staticmethod def count_flops(model, _x, y): return count_flops_attn(model, _x, y) class ConditionalVideoUNet(nn.Module): def __init__( self, image_size, in_channels, model_channels, out_channels, num_res_blocks, attention_resolutions, dropout=0, channel_mult=(1, 2, 4, 8), conv_resample=True, dims=2, num_frames=1, num_classes=None, 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, transformer_depth=1, context_dim=None, n_embed=None, legacy=True, disable_self_attentions=None, num_attention_blocks=None, disable_middle_self_attn=False, use_linear_in_transformer=False, condition_channels=None ): 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.image_size = image_size self.num_frames = num_frames self.in_channels = in_channels self.model_channels = model_channels self.out_channels = out_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: 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.num_classes = num_classes 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), ) if self.num_classes is not None: if isinstance(self.num_classes, int): self.label_emb = nn.Embedding(num_classes, time_embed_dim) elif self.num_classes == "continuous": print("setting up linear c_adm embedding layer") self.label_emb = nn.Linear(1, time_embed_dim) else: raise ValueError() self.input_blocks = nn.ModuleList( [ TimestepEmbedSequential( conv_nd(dims, in_channels, 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 = [ PriorConditionedResBlock( ch, time_embed_dim, dropout, condition_channels=condition_channels, 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: 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 ProjectedSpatialTransformer( 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( PriorConditionedResBlock( ch, time_embed_dim, dropout, condition_channels=condition_channels, 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: dim_head = ch // num_heads if use_spatial_transformer else num_head_channels self.middle_block = TimestepEmbedSequential( PriorConditionedResBlock( ch, time_embed_dim, dropout, condition_channels=condition_channels, 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 ProjectedSpatialTransformer( 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 ), PriorConditionedResBlock( ch, time_embed_dim, dropout, condition_channels=condition_channels, dims=dims, use_checkpoint=use_checkpoint, use_scale_shift_norm=use_scale_shift_norm, ), ) self.middle_temporal_conv0 = TimestepEmbedSequential(SpatialTemporalConv(mult * model_channels, num_frames=num_frames)) self.middle_temporal_attn = TimestepEmbedSequential(TemporalAttention(ch,num_heads=num_heads,dim_head=dim_head,num_frames=num_frames)) self.middle_temporal_conv1 = TimestepEmbedSequential(SpatialTemporalConv(mult * model_channels, num_frames=num_frames)) self._feature_size += ch self.output_blocks = nn.ModuleList([]) for level, mult in list(enumerate(channel_mult))[::-1]: for i in range(self.num_res_blocks[level] + 1): ich = input_block_chans.pop() layers = [ PriorConditionedResBlock( ch + ich, time_embed_dim, dropout, condition_channels=condition_channels, out_channels=model_channels * mult, dims=dims, use_checkpoint=use_checkpoint, use_scale_shift_norm=use_scale_shift_norm, ), ] ch = model_channels * mult 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: 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 i < num_attention_blocks[level]: layers.append( AttentionBlock( ch, use_checkpoint=use_checkpoint, num_heads=num_heads_upsample, num_head_channels=dim_head, use_new_attention_order=use_new_attention_order, ) if not use_spatial_transformer else ProjectedSpatialTransformer( 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 ) ) if level and i == self.num_res_blocks[level]: out_ch = ch layers.append( PriorConditionedResBlock( ch, time_embed_dim, dropout, condition_channels=condition_channels, out_channels=out_ch, dims=dims, use_checkpoint=use_checkpoint, use_scale_shift_norm=use_scale_shift_norm, up=True, ) if resblock_updown else Upsample(ch, conv_resample, dims=dims, out_channels=out_ch) ) ds //= 2 self.output_blocks.append(TimestepEmbedSequential(*layers)) self._feature_size += ch self.out = nn.Sequential( normalization(ch), nn.SiLU(), zero_module(conv_nd(dims, model_channels, out_channels, 3, padding=1)), ) if self.predict_codebook_ids: self.id_predictor = nn.Sequential( normalization(ch), conv_nd(dims, model_channels, n_embed, 1), ) def convert_to_fp16(self): self.input_blocks.apply(convert_module_to_f16) self.middle_block.apply(convert_module_to_f16) self.output_blocks.apply(convert_module_to_f16) def convert_to_fp32(self): self.input_blocks.apply(convert_module_to_f32) self.middle_block.apply(convert_module_to_f32) self.output_blocks.apply(convert_module_to_f32) def forward(self, x, timesteps=None, context=None, struct_cond=None, prior_cond=None, y=None, **kwargs): assert (y is not None) == ( self.num_classes is not None ), "must specify y if and only if the model is class-conditional" hs = [] t_emb = timestep_embedding(timesteps, self.model_channels, repeat_only=False) emb = self.time_embed(t_emb) if self.num_classes is not None: assert y.shape == (x.shape[0],) emb = emb + self.label_emb(y) h = x.type(self.dtype) for module in self.input_blocks: h = module(h, emb, context, struct_cond, prior_cond) hs.append(h) h = self.middle_block[0](h, emb, struct_cond, prior_cond) h = self.middle_temporal_conv0(h, emb, context, struct_cond) h = self.middle_block[1](h, context) h = self.middle_temporal_attn(h, emb, context, struct_cond) h = self.middle_block[2](h, emb, struct_cond, prior_cond) h = self.middle_temporal_conv1(h, emb, context, struct_cond) for module in self.output_blocks: h = th.cat([h, hs.pop()], dim=1) h = module(h, emb, context, struct_cond, prior_cond) h = h.type(x.dtype) if self.predict_codebook_ids: return self.id_predictor(h) else: return self.out(h) class PriorConditionEncoder(nn.Module): def __init__( self, prior_size=32, in_channels=256, model_channels=256, out_channels=256, num_res_blocks=1, dropout= 0, channel_mult=(1, 1, 1, 1), conv_resample=True, dims=2, use_checkpoint=False, use_fp16=False, use_scale_shift_norm=False, resblock_updown=False, *args, **kwargs ): super().__init__() self.in_channels = in_channels self.model_channels = model_channels self.out_channels = out_channels self.num_res_blocks = num_res_blocks self.dropout = dropout self.channel_mult = channel_mult self.conv_resample = conv_resample self.use_checkpoint = use_checkpoint self.dtype = torch.float16 if use_fp16 else torch.float32 self.use_scale_shift_norm = use_scale_shift_norm time_embed_dim = model_channels * 4 self.time_embed = nn.Sequential( nn.Linear(model_channels, time_embed_dim), nn.SiLU(), nn.Linear(time_embed_dim, time_embed_dim), ) self.input_blocks = nn.ModuleList([ TimestepEmbedSequential( conv_nd(dims, in_channels, model_channels, 3, padding=1) ) ]) input_block_chans = [] ch = model_channels ds = 1 for level, mult in enumerate(channel_mult): for _ in range(num_res_blocks): 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 self.input_blocks.append(TimestepEmbedSequential(*layers)) 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.middle_block = TimestepEmbedSequential( ResBlock( ch, time_embed_dim, dropout, dims=dims, use_checkpoint=use_checkpoint, use_scale_shift_norm=use_scale_shift_norm, ) ) input_block_chans.append(ch) self.input_block_chans = input_block_chans self.cond_fea_tran = nn.ModuleList([]) self.diffusion_fea_tran = nn.ModuleList([]) for i in range(len(input_block_chans)): self.cond_fea_tran.append( ResBlock( input_block_chans[i], time_embed_dim, dropout, out_channels=out_channels, dims=dims, use_checkpoint=use_checkpoint, use_scale_shift_norm=use_scale_shift_norm, ) ) self.diffusion_fea_tran.append( ResBlock( input_block_chans[i], time_embed_dim, dropout, out_channels=out_channels, dims=dims, use_checkpoint=use_checkpoint, use_scale_shift_norm=use_scale_shift_norm, ) ) def forward(self, x, timesteps): emb = self.time_embed(timestep_embedding(timesteps, self.model_channels)) result_list = [] cond_results = {} diffusion_results = {} h = x.type(self.dtype) for module in self.input_blocks: last_h = h h = module(h, emb) if h.size(-1) != last_h.size(-1): result_list.append(last_h) h = self.middle_block(h, emb) result_list.append(h) assert len(result_list) == len(self.cond_fea_tran) assert len(result_list) == len(self.diffusion_fea_tran) for i in range(len(result_list)): key = str(result_list[i].size(-1)) cond_results[key] = self.cond_fea_tran[i](result_list[i], emb) diffusion_results[key] = self.diffusion_fea_tran[i](result_list[i], emb) return cond_results, diffusion_results class VideoStructureEncoder(nn.Module): def __init__( self, image_size, in_channels, model_channels, out_channels, num_res_blocks, attention_resolutions, dropout=0, channel_mult=(1, 2, 4, 8), conv_resample=True, dims=2, num_frames=1, 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, condition_channels=None, *args, **kwargs ): super().__init__() if num_heads_upsample == -1: num_heads_upsample = num_heads self.in_channels = in_channels self.model_channels = model_channels self.out_channels = out_channels self.num_res_blocks = num_res_blocks 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 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._feature_size = model_channels input_block_chans = [] ch = model_channels ds = 1 for level, mult in enumerate(channel_mult): for _ in range(num_res_blocks): layers = [ StructureConditionedResBlock( ch, time_embed_dim, dropout, condition_channels=condition_channels, 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: layers.append( AttentionBlock( ch, use_checkpoint=use_checkpoint, num_heads=num_heads, num_head_channels=num_head_channels, use_new_attention_order=use_new_attention_order, ) ) self.input_blocks.append(TimestepEmbedSequential(*layers)) self._feature_size += ch if level != len(channel_mult) - 1: out_ch = ch self.input_blocks.append( TimestepEmbedSequential( StructureConditionedResBlock( ch, time_embed_dim, dropout, condition_channels=condition_channels, 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 self.middle_block = TimestepEmbedSequential( StructureConditionedResBlock( ch, time_embed_dim, dropout, condition_channels=condition_channels, 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=num_head_channels, use_new_attention_order=use_new_attention_order, ), StructureConditionedResBlock( ch, time_embed_dim, dropout, condition_channels=condition_channels, dims=dims, use_checkpoint=use_checkpoint, use_scale_shift_norm=use_scale_shift_norm, ), ) input_block_chans.append(ch) self._feature_size += ch self.input_block_chans = input_block_chans self.fea_tran = nn.ModuleList([]) for i in range(len(input_block_chans)): self.fea_tran.append( StructureConditionedResBlock( input_block_chans[i], time_embed_dim, dropout, condition_channels=condition_channels, out_channels=out_channels, dims=dims, use_checkpoint=use_checkpoint, use_scale_shift_norm=use_scale_shift_norm, ) ) def convert_to_fp16(self): self.input_blocks.apply(convert_module_to_f16) self.middle_block.apply(convert_module_to_f16) def convert_to_fp32(self): self.input_blocks.apply(convert_module_to_f32) self.middle_block.apply(convert_module_to_f32) def forward(self, x, timesteps, prior): emb = self.time_embed(timestep_embedding(timesteps, self.model_channels)) result_list = [] results = {} h = x.type(self.dtype) for module in self.input_blocks: last_h = h h = module(h, emb, struct_cond=prior) if h.size(-1) != last_h.size(-1): result_list.append(last_h) h = self.middle_block(h, emb, struct_cond=prior) result_list.append(h) assert len(result_list) == len(self.fea_tran) for i in range(len(result_list)): results[str(result_list[i].size(-1))] = self.fea_tran[i](result_list[i], emb, s_cond=prior) return results