Spaces:
Runtime error
Runtime error
BestWishYsh
commited on
Update models/transformer_consisid.py
Browse files- models/transformer_consisid.py +81 -227
models/transformer_consisid.py
CHANGED
@@ -16,7 +16,7 @@ import glob
|
|
16 |
import json
|
17 |
import math
|
18 |
import os
|
19 |
-
from typing import Any, Dict, Optional, Tuple, Union
|
20 |
|
21 |
import torch
|
22 |
from torch import nn
|
@@ -24,11 +24,7 @@ from torch import nn
|
|
24 |
from diffusers.configuration_utils import ConfigMixin, register_to_config
|
25 |
from diffusers.loaders import PeftAdapterMixin
|
26 |
from diffusers.models.attention import Attention, FeedForward
|
27 |
-
from diffusers.models.attention_processor import
|
28 |
-
AttentionProcessor,
|
29 |
-
CogVideoXAttnProcessor2_0,
|
30 |
-
FusedCogVideoXAttnProcessor2_0,
|
31 |
-
)
|
32 |
from diffusers.models.embeddings import CogVideoXPatchEmbed, TimestepEmbedding, Timesteps
|
33 |
from diffusers.models.modeling_outputs import Transformer2DModelOutput
|
34 |
from diffusers.models.modeling_utils import ModelMixin
|
@@ -40,61 +36,10 @@ from diffusers.utils.torch_utils import maybe_allow_in_graph
|
|
40 |
logger = logging.get_logger(__name__) # pylint: disable=invalid-name
|
41 |
|
42 |
|
43 |
-
def ConsisIDFeedForward(dim, mult=4):
|
44 |
-
"""
|
45 |
-
Creates a consistent ID feedforward block consisting of layer normalization, two linear layers, and a GELU
|
46 |
-
activation.
|
47 |
-
|
48 |
-
Args:
|
49 |
-
dim (int): The input dimension of the tensor.
|
50 |
-
mult (int, optional): Multiplier for the inner dimension. Default is 4.
|
51 |
-
|
52 |
-
Returns:
|
53 |
-
nn.Sequential: A sequence of layers comprising LayerNorm, Linear layers, and GELU.
|
54 |
-
"""
|
55 |
-
inner_dim = int(dim * mult)
|
56 |
-
return nn.Sequential(
|
57 |
-
nn.LayerNorm(dim),
|
58 |
-
nn.Linear(dim, inner_dim, bias=False),
|
59 |
-
nn.GELU(),
|
60 |
-
nn.Linear(inner_dim, dim, bias=False),
|
61 |
-
)
|
62 |
-
|
63 |
-
|
64 |
-
def reshape_tensor(x, heads):
|
65 |
-
"""
|
66 |
-
Reshapes the input tensor for multi-head attention.
|
67 |
-
|
68 |
-
Args:
|
69 |
-
x (torch.Tensor): The input tensor with shape (batch_size, length, width).
|
70 |
-
heads (int): The number of attention heads.
|
71 |
-
|
72 |
-
Returns:
|
73 |
-
torch.Tensor: The reshaped tensor, with shape (batch_size, heads, length, width).
|
74 |
-
"""
|
75 |
-
bs, length, width = x.shape
|
76 |
-
x = x.view(bs, length, heads, -1)
|
77 |
-
x = x.transpose(1, 2)
|
78 |
-
x = x.reshape(bs, heads, length, -1)
|
79 |
-
return x
|
80 |
-
|
81 |
-
|
82 |
class PerceiverAttention(nn.Module):
|
83 |
-
|
84 |
-
Implements the Perceiver attention mechanism with multi-head attention.
|
85 |
-
|
86 |
-
This layer takes two inputs: 'x' (image features) and 'latents' (latent features), applying multi-head attention to
|
87 |
-
both and producing an output tensor with the same dimension as the input tensor 'x'.
|
88 |
-
|
89 |
-
Args:
|
90 |
-
dim (int): The input dimension.
|
91 |
-
dim_head (int, optional): The dimension of each attention head. Default is 64.
|
92 |
-
heads (int, optional): The number of attention heads. Default is 8.
|
93 |
-
kv_dim (int, optional): The key-value dimension. If None, `dim` is used for both keys and values.
|
94 |
-
"""
|
95 |
-
|
96 |
-
def __init__(self, *, dim, dim_head=64, heads=8, kv_dim=None):
|
97 |
super().__init__()
|
|
|
98 |
self.scale = dim_head**-0.5
|
99 |
self.dim_head = dim_head
|
100 |
self.heads = heads
|
@@ -107,80 +52,58 @@ class PerceiverAttention(nn.Module):
|
|
107 |
self.to_kv = nn.Linear(dim if kv_dim is None else kv_dim, inner_dim * 2, bias=False)
|
108 |
self.to_out = nn.Linear(inner_dim, dim, bias=False)
|
109 |
|
110 |
-
def forward(self,
|
111 |
-
"""
|
112 |
-
Forward pass for Perceiver attention.
|
113 |
-
|
114 |
-
Args:
|
115 |
-
x (torch.Tensor): Image features tensor with shape (batch_size, num_pixels, D).
|
116 |
-
latents (torch.Tensor): Latent features tensor with shape (batch_size, num_latents, D).
|
117 |
-
|
118 |
-
Returns:
|
119 |
-
torch.Tensor: Output tensor after applying attention and transformation.
|
120 |
-
"""
|
121 |
# Apply normalization
|
122 |
-
|
123 |
latents = self.norm2(latents)
|
124 |
|
125 |
-
|
126 |
|
127 |
# Compute query, key, and value matrices
|
128 |
-
|
129 |
-
kv_input = torch.cat((
|
130 |
-
|
131 |
|
132 |
# Reshape the tensors for multi-head attention
|
133 |
-
|
134 |
-
|
135 |
-
|
136 |
|
137 |
# attention
|
138 |
scale = 1 / math.sqrt(math.sqrt(self.dim_head))
|
139 |
-
weight = (
|
140 |
weight = torch.softmax(weight.float(), dim=-1).type(weight.dtype)
|
141 |
-
|
142 |
|
143 |
# Reshape and return the final output
|
144 |
-
|
145 |
|
146 |
-
return self.to_out(
|
147 |
|
148 |
|
149 |
class LocalFacialExtractor(nn.Module):
|
150 |
def __init__(
|
151 |
self,
|
152 |
-
id_dim=1280,
|
153 |
-
vit_dim=1024,
|
154 |
-
depth=10,
|
155 |
-
dim_head=64,
|
156 |
-
heads=16,
|
157 |
-
num_id_token=5,
|
158 |
-
num_queries=32,
|
159 |
-
output_dim=2048,
|
160 |
-
ff_mult=4,
|
|
|
161 |
):
|
162 |
-
"""
|
163 |
-
Initializes the LocalFacialExtractor class.
|
164 |
-
|
165 |
-
Parameters:
|
166 |
-
- id_dim (int): The dimensionality of id features.
|
167 |
-
- vit_dim (int): The dimensionality of vit features.
|
168 |
-
- depth (int): Total number of PerceiverAttention and ConsisIDFeedForward layers.
|
169 |
-
- dim_head (int): Dimensionality of each attention head.
|
170 |
-
- heads (int): Number of attention heads.
|
171 |
-
- num_id_token (int): Number of tokens used for identity features.
|
172 |
-
- num_queries (int): Number of query tokens for the latent representation.
|
173 |
-
- output_dim (int): Output dimension after projection.
|
174 |
-
- ff_mult (int): Multiplier for the feed-forward network hidden dimension.
|
175 |
-
"""
|
176 |
super().__init__()
|
177 |
|
178 |
# Storing identity token and query information
|
179 |
self.num_id_token = num_id_token
|
180 |
self.vit_dim = vit_dim
|
181 |
self.num_queries = num_queries
|
182 |
-
assert depth %
|
183 |
-
self.depth = depth //
|
|
|
184 |
scale = vit_dim**-0.5
|
185 |
|
186 |
# Learnable latent query embeddings
|
@@ -195,13 +118,18 @@ class LocalFacialExtractor(nn.Module):
|
|
195 |
nn.ModuleList(
|
196 |
[
|
197 |
PerceiverAttention(dim=vit_dim, dim_head=dim_head, heads=heads), # Perceiver Attention layer
|
198 |
-
|
|
|
|
|
|
|
|
|
|
|
199 |
]
|
200 |
)
|
201 |
)
|
202 |
|
203 |
# Mappings for each of the 5 different ViT features
|
204 |
-
for i in range(
|
205 |
setattr(
|
206 |
self,
|
207 |
f"mapping_{i}",
|
@@ -227,32 +155,21 @@ class LocalFacialExtractor(nn.Module):
|
|
227 |
nn.Linear(vit_dim, vit_dim * num_id_token),
|
228 |
)
|
229 |
|
230 |
-
def forward(self,
|
231 |
-
"""
|
232 |
-
Forward pass for LocalFacialExtractor.
|
233 |
-
|
234 |
-
Parameters:
|
235 |
-
- x (Tensor): The input identity embedding tensor of shape (batch_size, id_dim).
|
236 |
-
- y (list of Tensor): A list of 5 visual feature tensors each of shape (batch_size, vit_dim).
|
237 |
-
|
238 |
-
Returns:
|
239 |
-
- Tensor: The extracted latent features of shape (batch_size, num_queries, output_dim).
|
240 |
-
"""
|
241 |
-
|
242 |
# Repeat latent queries for the batch size
|
243 |
-
latents = self.latents.repeat(
|
244 |
|
245 |
# Map the identity embedding to tokens
|
246 |
-
|
247 |
-
|
248 |
|
249 |
# Concatenate identity tokens with the latent queries
|
250 |
-
latents = torch.cat((latents,
|
251 |
|
252 |
-
# Process each of the
|
253 |
-
for i in range(
|
254 |
-
vit_feature = getattr(self, f"mapping_{i}")(
|
255 |
-
ctx_feature = torch.cat((
|
256 |
|
257 |
# Pass through the PerceiverAttention and ConsisIDFeedForward layers
|
258 |
for attn, ff in self.layers[i * self.depth : (i + 1) * self.depth]:
|
@@ -267,26 +184,9 @@ class LocalFacialExtractor(nn.Module):
|
|
267 |
|
268 |
|
269 |
class PerceiverCrossAttention(nn.Module):
|
270 |
-
|
271 |
-
|
272 |
-
Args:
|
273 |
-
dim (int): Dimension of the input latent and output. Default is 3072.
|
274 |
-
dim_head (int): Dimension of each attention head. Default is 128.
|
275 |
-
heads (int): Number of attention heads. Default is 16.
|
276 |
-
kv_dim (int): Dimension of the key/value input, allowing flexible cross-attention. Default is 2048.
|
277 |
-
|
278 |
-
Attributes:
|
279 |
-
scale (float): Scaling factor used in dot-product attention for numerical stability.
|
280 |
-
norm1 (nn.LayerNorm): Layer normalization applied to the input image features.
|
281 |
-
norm2 (nn.LayerNorm): Layer normalization applied to the latent features.
|
282 |
-
to_q (nn.Linear): Linear layer for projecting the latent features into queries.
|
283 |
-
to_kv (nn.Linear): Linear layer for projecting the input features into keys and values.
|
284 |
-
to_out (nn.Linear): Linear layer for outputting the final result after attention.
|
285 |
-
|
286 |
-
"""
|
287 |
-
|
288 |
-
def __init__(self, *, dim=3072, dim_head=128, heads=16, kv_dim=2048):
|
289 |
super().__init__()
|
|
|
290 |
self.scale = dim_head**-0.5
|
291 |
self.dim_head = dim_head
|
292 |
self.heads = heads
|
@@ -301,47 +201,32 @@ class PerceiverCrossAttention(nn.Module):
|
|
301 |
self.to_kv = nn.Linear(dim if kv_dim is None else kv_dim, inner_dim * 2, bias=False)
|
302 |
self.to_out = nn.Linear(inner_dim, dim, bias=False)
|
303 |
|
304 |
-
def forward(self,
|
305 |
-
"""
|
306 |
-
|
307 |
-
Args:
|
308 |
-
x (torch.Tensor): Input image features with shape (batch_size, n1, D), where:
|
309 |
-
- batch_size (b): Number of samples in the batch.
|
310 |
-
- n1: Sequence length (e.g., number of patches or tokens).
|
311 |
-
- D: Feature dimension.
|
312 |
-
|
313 |
-
latents (torch.Tensor): Latent feature representations with shape (batch_size, n2, D), where:
|
314 |
-
- n2: Number of latent elements.
|
315 |
-
|
316 |
-
Returns:
|
317 |
-
torch.Tensor: Attention-modulated features with shape (batch_size, n2, D).
|
318 |
-
|
319 |
-
"""
|
320 |
# Apply layer normalization to the input image and latent features
|
321 |
-
|
322 |
-
|
323 |
|
324 |
-
|
325 |
|
326 |
# Compute queries, keys, and values
|
327 |
-
|
328 |
-
|
329 |
|
330 |
# Reshape tensors to split into attention heads
|
331 |
-
|
332 |
-
|
333 |
-
|
334 |
|
335 |
# Compute attention weights
|
336 |
scale = 1 / math.sqrt(math.sqrt(self.dim_head))
|
337 |
-
weight = (
|
338 |
weight = torch.softmax(weight.float(), dim=-1).type(weight.dtype)
|
339 |
|
340 |
# Compute the output via weighted combination of values
|
341 |
-
out = weight @
|
342 |
|
343 |
# Reshape and permute to prepare for final linear transformation
|
344 |
-
out = out.permute(0, 2, 1, 3).reshape(
|
345 |
|
346 |
return self.to_out(out)
|
347 |
|
@@ -567,6 +452,9 @@ class ConsisIDTransformer3DModel(ModelMixin, ConfigMixin, PeftAdapterMixin):
|
|
567 |
The multiplication factor applied to the feed-forward network's hidden layer size in the Local Facial
|
568 |
Extractor (LFE). A higher value increases the model's capacity to learn more complex facial feature
|
569 |
transformations, but also increases the computation and memory requirements.
|
|
|
|
|
|
|
570 |
local_face_scale (`float`, defaults to `1.0`):
|
571 |
A scaling factor used to adjust the importance of local facial features in the model. This can influence
|
572 |
how strongly the model focuses on high frequency face-related content.
|
@@ -616,6 +504,7 @@ class ConsisIDTransformer3DModel(ModelMixin, ConfigMixin, PeftAdapterMixin):
|
|
616 |
LFE_num_querie: int = 32,
|
617 |
LFE_output_dim: int = 2048,
|
618 |
LFE_ff_mult: int = 4,
|
|
|
619 |
local_face_scale: float = 1.0,
|
620 |
):
|
621 |
super().__init__()
|
@@ -680,8 +569,6 @@ class ConsisIDTransformer3DModel(ModelMixin, ConfigMixin, PeftAdapterMixin):
|
|
680 |
)
|
681 |
self.proj_out = nn.Linear(inner_dim, patch_size * patch_size * out_channels)
|
682 |
|
683 |
-
self.gradient_checkpointing = False
|
684 |
-
|
685 |
self.is_train_face = is_train_face
|
686 |
self.is_kps = is_kps
|
687 |
|
@@ -697,6 +584,7 @@ class ConsisIDTransformer3DModel(ModelMixin, ConfigMixin, PeftAdapterMixin):
|
|
697 |
self.LFE_num_querie = LFE_num_querie
|
698 |
self.LFE_output_dim = LFE_output_dim
|
699 |
self.LFE_ff_mult = LFE_ff_mult
|
|
|
700 |
# cross configs
|
701 |
self.inner_dim = inner_dim
|
702 |
self.cross_attn_interval = cross_attn_interval
|
@@ -708,6 +596,8 @@ class ConsisIDTransformer3DModel(ModelMixin, ConfigMixin, PeftAdapterMixin):
|
|
708 |
# face modules
|
709 |
self._init_face_inputs()
|
710 |
|
|
|
|
|
711 |
def _set_gradient_checkpointing(self, module, value=False):
|
712 |
self.gradient_checkpointing = value
|
713 |
|
@@ -724,8 +614,8 @@ class ConsisIDTransformer3DModel(ModelMixin, ConfigMixin, PeftAdapterMixin):
|
|
724 |
num_queries=self.LFE_num_querie,
|
725 |
output_dim=self.LFE_output_dim,
|
726 |
ff_mult=self.LFE_ff_mult,
|
727 |
-
|
728 |
-
|
729 |
self.perceiver_cross_attention = nn.ModuleList(
|
730 |
[
|
731 |
PerceiverCrossAttention(
|
@@ -811,46 +701,6 @@ class ConsisIDTransformer3DModel(ModelMixin, ConfigMixin, PeftAdapterMixin):
|
|
811 |
for name, module in self.named_children():
|
812 |
fn_recursive_attn_processor(name, module, processor)
|
813 |
|
814 |
-
# Copied from diffusers.models.unets.unet_2d_condition.UNet2DConditionModel.fuse_qkv_projections with FusedAttnProcessor2_0->FusedCogVideoXAttnProcessor2_0
|
815 |
-
def fuse_qkv_projections(self):
|
816 |
-
"""
|
817 |
-
Enables fused QKV projections. For self-attention modules, all projection matrices (i.e., query, key, value)
|
818 |
-
are fused. For cross-attention modules, key and value projection matrices are fused.
|
819 |
-
|
820 |
-
<Tip warning={true}>
|
821 |
-
|
822 |
-
This API is 🧪 experimental.
|
823 |
-
|
824 |
-
</Tip>
|
825 |
-
"""
|
826 |
-
self.original_attn_processors = None
|
827 |
-
|
828 |
-
for _, attn_processor in self.attn_processors.items():
|
829 |
-
if "Added" in str(attn_processor.__class__.__name__):
|
830 |
-
raise ValueError("`fuse_qkv_projections()` is not supported for models having added KV projections.")
|
831 |
-
|
832 |
-
self.original_attn_processors = self.attn_processors
|
833 |
-
|
834 |
-
for module in self.modules():
|
835 |
-
if isinstance(module, Attention):
|
836 |
-
module.fuse_projections(fuse=True)
|
837 |
-
|
838 |
-
self.set_attn_processor(FusedCogVideoXAttnProcessor2_0())
|
839 |
-
|
840 |
-
# Copied from diffusers.models.unets.unet_2d_condition.UNet2DConditionModel.unfuse_qkv_projections
|
841 |
-
def unfuse_qkv_projections(self):
|
842 |
-
"""Disables the fused QKV projection if enabled.
|
843 |
-
|
844 |
-
<Tip warning={true}>
|
845 |
-
|
846 |
-
This API is 🧪 experimental.
|
847 |
-
|
848 |
-
</Tip>
|
849 |
-
|
850 |
-
"""
|
851 |
-
if self.original_attn_processors is not None:
|
852 |
-
self.set_attn_processor(self.original_attn_processors)
|
853 |
-
|
854 |
def forward(
|
855 |
self,
|
856 |
hidden_states: torch.Tensor,
|
@@ -863,13 +713,6 @@ class ConsisIDTransformer3DModel(ModelMixin, ConfigMixin, PeftAdapterMixin):
|
|
863 |
id_vit_hidden: Optional[torch.Tensor] = None,
|
864 |
return_dict: bool = True,
|
865 |
):
|
866 |
-
# fuse clip and insightface
|
867 |
-
if self.is_train_face:
|
868 |
-
assert id_cond is not None and id_vit_hidden is not None
|
869 |
-
valid_face_emb = self.local_facial_extractor(
|
870 |
-
id_cond, id_vit_hidden
|
871 |
-
) # torch.Size([1, 1280]), list[5](torch.Size([1, 577, 1024])) -> torch.Size([1, 32, 2048])
|
872 |
-
|
873 |
if attention_kwargs is not None:
|
874 |
attention_kwargs = attention_kwargs.copy()
|
875 |
lora_scale = attention_kwargs.pop("scale", 1.0)
|
@@ -885,6 +728,17 @@ class ConsisIDTransformer3DModel(ModelMixin, ConfigMixin, PeftAdapterMixin):
|
|
885 |
"Passing `scale` via `attention_kwargs` when not using the PEFT backend is ineffective."
|
886 |
)
|
887 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
888 |
batch_size, num_frames, channels, height, width = hidden_states.shape
|
889 |
|
890 |
# 1. Time embedding
|
@@ -1086,4 +940,4 @@ if __name__ == '__main__':
|
|
1086 |
id_cond=id_cond if id_cond is not None else None,
|
1087 |
)[0]
|
1088 |
|
1089 |
-
print(model_output)
|
|
|
16 |
import json
|
17 |
import math
|
18 |
import os
|
19 |
+
from typing import Any, List, Dict, Optional, Tuple, Union
|
20 |
|
21 |
import torch
|
22 |
from torch import nn
|
|
|
24 |
from diffusers.configuration_utils import ConfigMixin, register_to_config
|
25 |
from diffusers.loaders import PeftAdapterMixin
|
26 |
from diffusers.models.attention import Attention, FeedForward
|
27 |
+
from diffusers.models.attention_processor import AttentionProcessor, CogVideoXAttnProcessor2_0
|
|
|
|
|
|
|
|
|
28 |
from diffusers.models.embeddings import CogVideoXPatchEmbed, TimestepEmbedding, Timesteps
|
29 |
from diffusers.models.modeling_outputs import Transformer2DModelOutput
|
30 |
from diffusers.models.modeling_utils import ModelMixin
|
|
|
36 |
logger = logging.get_logger(__name__) # pylint: disable=invalid-name
|
37 |
|
38 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
39 |
class PerceiverAttention(nn.Module):
|
40 |
+
def __init__(self, dim: int, dim_head: int = 64, heads: int = 8, kv_dim: Optional[int] = None):
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
41 |
super().__init__()
|
42 |
+
|
43 |
self.scale = dim_head**-0.5
|
44 |
self.dim_head = dim_head
|
45 |
self.heads = heads
|
|
|
52 |
self.to_kv = nn.Linear(dim if kv_dim is None else kv_dim, inner_dim * 2, bias=False)
|
53 |
self.to_out = nn.Linear(inner_dim, dim, bias=False)
|
54 |
|
55 |
+
def forward(self, image_embeds: torch.Tensor, latents: torch.Tensor) -> torch.Tensor:
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
56 |
# Apply normalization
|
57 |
+
image_embeds = self.norm1(image_embeds)
|
58 |
latents = self.norm2(latents)
|
59 |
|
60 |
+
batch_size, seq_len, _ = latents.shape # Get batch size and sequence length
|
61 |
|
62 |
# Compute query, key, and value matrices
|
63 |
+
query = self.to_q(latents)
|
64 |
+
kv_input = torch.cat((image_embeds, latents), dim=-2)
|
65 |
+
key, value = self.to_kv(kv_input).chunk(2, dim=-1)
|
66 |
|
67 |
# Reshape the tensors for multi-head attention
|
68 |
+
query = query.reshape(query.size(0), -1, self.heads, self.dim_head).transpose(1, 2)
|
69 |
+
key = key.reshape(key.size(0), -1, self.heads, self.dim_head).transpose(1, 2)
|
70 |
+
value = value.reshape(value.size(0), -1, self.heads, self.dim_head).transpose(1, 2)
|
71 |
|
72 |
# attention
|
73 |
scale = 1 / math.sqrt(math.sqrt(self.dim_head))
|
74 |
+
weight = (query * scale) @ (key * scale).transpose(-2, -1) # More stable with f16 than dividing afterwards
|
75 |
weight = torch.softmax(weight.float(), dim=-1).type(weight.dtype)
|
76 |
+
output = weight @ value
|
77 |
|
78 |
# Reshape and return the final output
|
79 |
+
output = output.permute(0, 2, 1, 3).reshape(batch_size, seq_len, -1)
|
80 |
|
81 |
+
return self.to_out(output)
|
82 |
|
83 |
|
84 |
class LocalFacialExtractor(nn.Module):
|
85 |
def __init__(
|
86 |
self,
|
87 |
+
id_dim: int = 1280,
|
88 |
+
vit_dim: int = 1024,
|
89 |
+
depth: int = 10,
|
90 |
+
dim_head: int = 64,
|
91 |
+
heads: int = 16,
|
92 |
+
num_id_token: int = 5,
|
93 |
+
num_queries: int = 32,
|
94 |
+
output_dim: int = 2048,
|
95 |
+
ff_mult: int = 4,
|
96 |
+
num_scale: int = 5,
|
97 |
):
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
98 |
super().__init__()
|
99 |
|
100 |
# Storing identity token and query information
|
101 |
self.num_id_token = num_id_token
|
102 |
self.vit_dim = vit_dim
|
103 |
self.num_queries = num_queries
|
104 |
+
assert depth % num_scale == 0
|
105 |
+
self.depth = depth // num_scale
|
106 |
+
self.num_scale = num_scale
|
107 |
scale = vit_dim**-0.5
|
108 |
|
109 |
# Learnable latent query embeddings
|
|
|
118 |
nn.ModuleList(
|
119 |
[
|
120 |
PerceiverAttention(dim=vit_dim, dim_head=dim_head, heads=heads), # Perceiver Attention layer
|
121 |
+
nn.Sequential(
|
122 |
+
nn.LayerNorm(vit_dim),
|
123 |
+
nn.Linear(vit_dim, vit_dim * ff_mult, bias=False),
|
124 |
+
nn.GELU(),
|
125 |
+
nn.Linear(vit_dim * ff_mult, vit_dim, bias=False),
|
126 |
+
), # ConsisIDFeedForward layer
|
127 |
]
|
128 |
)
|
129 |
)
|
130 |
|
131 |
# Mappings for each of the 5 different ViT features
|
132 |
+
for i in range(num_scale):
|
133 |
setattr(
|
134 |
self,
|
135 |
f"mapping_{i}",
|
|
|
155 |
nn.Linear(vit_dim, vit_dim * num_id_token),
|
156 |
)
|
157 |
|
158 |
+
def forward(self, id_embeds: torch.Tensor, vit_hidden_states: List[torch.Tensor]) -> torch.Tensor:
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
159 |
# Repeat latent queries for the batch size
|
160 |
+
latents = self.latents.repeat(id_embeds.size(0), 1, 1)
|
161 |
|
162 |
# Map the identity embedding to tokens
|
163 |
+
id_embeds = self.id_embedding_mapping(id_embeds)
|
164 |
+
id_embeds = id_embeds.reshape(-1, self.num_id_token, self.vit_dim)
|
165 |
|
166 |
# Concatenate identity tokens with the latent queries
|
167 |
+
latents = torch.cat((latents, id_embeds), dim=1)
|
168 |
|
169 |
+
# Process each of the num_scale visual feature inputs
|
170 |
+
for i in range(self.num_scale):
|
171 |
+
vit_feature = getattr(self, f"mapping_{i}")(vit_hidden_states[i])
|
172 |
+
ctx_feature = torch.cat((id_embeds, vit_feature), dim=1)
|
173 |
|
174 |
# Pass through the PerceiverAttention and ConsisIDFeedForward layers
|
175 |
for attn, ff in self.layers[i * self.depth : (i + 1) * self.depth]:
|
|
|
184 |
|
185 |
|
186 |
class PerceiverCrossAttention(nn.Module):
|
187 |
+
def __init__(self, dim: int = 3072, dim_head: int = 128, heads: int = 16, kv_dim: int = 2048):
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
188 |
super().__init__()
|
189 |
+
|
190 |
self.scale = dim_head**-0.5
|
191 |
self.dim_head = dim_head
|
192 |
self.heads = heads
|
|
|
201 |
self.to_kv = nn.Linear(dim if kv_dim is None else kv_dim, inner_dim * 2, bias=False)
|
202 |
self.to_out = nn.Linear(inner_dim, dim, bias=False)
|
203 |
|
204 |
+
def forward(self, image_embeds: torch.Tensor, hidden_states: torch.Tensor) -> torch.Tensor:
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
205 |
# Apply layer normalization to the input image and latent features
|
206 |
+
image_embeds = self.norm1(image_embeds)
|
207 |
+
hidden_states = self.norm2(hidden_states)
|
208 |
|
209 |
+
batch_size, seq_len, _ = hidden_states.shape
|
210 |
|
211 |
# Compute queries, keys, and values
|
212 |
+
query = self.to_q(hidden_states)
|
213 |
+
key, value = self.to_kv(image_embeds).chunk(2, dim=-1)
|
214 |
|
215 |
# Reshape tensors to split into attention heads
|
216 |
+
query = query.reshape(query.size(0), -1, self.heads, self.dim_head).transpose(1, 2)
|
217 |
+
key = key.reshape(key.size(0), -1, self.heads, self.dim_head).transpose(1, 2)
|
218 |
+
value = value.reshape(value.size(0), -1, self.heads, self.dim_head).transpose(1, 2)
|
219 |
|
220 |
# Compute attention weights
|
221 |
scale = 1 / math.sqrt(math.sqrt(self.dim_head))
|
222 |
+
weight = (query * scale) @ (key * scale).transpose(-2, -1) # More stable scaling than post-division
|
223 |
weight = torch.softmax(weight.float(), dim=-1).type(weight.dtype)
|
224 |
|
225 |
# Compute the output via weighted combination of values
|
226 |
+
out = weight @ value
|
227 |
|
228 |
# Reshape and permute to prepare for final linear transformation
|
229 |
+
out = out.permute(0, 2, 1, 3).reshape(batch_size, seq_len, -1)
|
230 |
|
231 |
return self.to_out(out)
|
232 |
|
|
|
452 |
The multiplication factor applied to the feed-forward network's hidden layer size in the Local Facial
|
453 |
Extractor (LFE). A higher value increases the model's capacity to learn more complex facial feature
|
454 |
transformations, but also increases the computation and memory requirements.
|
455 |
+
LFE_num_scale (`int`, optional, defaults to `5`):
|
456 |
+
The number of different scales visual feature. A higher value increases the model's capacity to learn more
|
457 |
+
complex facial feature transformations, but also increases the computation and memory requirements.
|
458 |
local_face_scale (`float`, defaults to `1.0`):
|
459 |
A scaling factor used to adjust the importance of local facial features in the model. This can influence
|
460 |
how strongly the model focuses on high frequency face-related content.
|
|
|
504 |
LFE_num_querie: int = 32,
|
505 |
LFE_output_dim: int = 2048,
|
506 |
LFE_ff_mult: int = 4,
|
507 |
+
LFE_num_scale: int = 5,
|
508 |
local_face_scale: float = 1.0,
|
509 |
):
|
510 |
super().__init__()
|
|
|
569 |
)
|
570 |
self.proj_out = nn.Linear(inner_dim, patch_size * patch_size * out_channels)
|
571 |
|
|
|
|
|
572 |
self.is_train_face = is_train_face
|
573 |
self.is_kps = is_kps
|
574 |
|
|
|
584 |
self.LFE_num_querie = LFE_num_querie
|
585 |
self.LFE_output_dim = LFE_output_dim
|
586 |
self.LFE_ff_mult = LFE_ff_mult
|
587 |
+
self.LFE_num_scale = LFE_num_scale
|
588 |
# cross configs
|
589 |
self.inner_dim = inner_dim
|
590 |
self.cross_attn_interval = cross_attn_interval
|
|
|
596 |
# face modules
|
597 |
self._init_face_inputs()
|
598 |
|
599 |
+
self.gradient_checkpointing = False
|
600 |
+
|
601 |
def _set_gradient_checkpointing(self, module, value=False):
|
602 |
self.gradient_checkpointing = value
|
603 |
|
|
|
614 |
num_queries=self.LFE_num_querie,
|
615 |
output_dim=self.LFE_output_dim,
|
616 |
ff_mult=self.LFE_ff_mult,
|
617 |
+
num_scale=self.LFE_num_scale,
|
618 |
+
).to(device, dtype=weight_dtype)
|
619 |
self.perceiver_cross_attention = nn.ModuleList(
|
620 |
[
|
621 |
PerceiverCrossAttention(
|
|
|
701 |
for name, module in self.named_children():
|
702 |
fn_recursive_attn_processor(name, module, processor)
|
703 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
704 |
def forward(
|
705 |
self,
|
706 |
hidden_states: torch.Tensor,
|
|
|
713 |
id_vit_hidden: Optional[torch.Tensor] = None,
|
714 |
return_dict: bool = True,
|
715 |
):
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
716 |
if attention_kwargs is not None:
|
717 |
attention_kwargs = attention_kwargs.copy()
|
718 |
lora_scale = attention_kwargs.pop("scale", 1.0)
|
|
|
728 |
"Passing `scale` via `attention_kwargs` when not using the PEFT backend is ineffective."
|
729 |
)
|
730 |
|
731 |
+
# fuse clip and insightface
|
732 |
+
valid_face_emb = None
|
733 |
+
if self.is_train_face:
|
734 |
+
id_cond = id_cond.to(device=hidden_states.device, dtype=hidden_states.dtype)
|
735 |
+
id_vit_hidden = [
|
736 |
+
tensor.to(device=hidden_states.device, dtype=hidden_states.dtype) for tensor in id_vit_hidden
|
737 |
+
]
|
738 |
+
valid_face_emb = self.local_facial_extractor(
|
739 |
+
id_cond, id_vit_hidden
|
740 |
+
) # torch.Size([1, 1280]), list[5](torch.Size([1, 577, 1024])) -> torch.Size([1, 32, 2048])
|
741 |
+
|
742 |
batch_size, num_frames, channels, height, width = hidden_states.shape
|
743 |
|
744 |
# 1. Time embedding
|
|
|
940 |
id_cond=id_cond if id_cond is not None else None,
|
941 |
)[0]
|
942 |
|
943 |
+
print(model_output)
|