-
Notifications
You must be signed in to change notification settings - Fork 133
/
faceformer.py
164 lines (153 loc) · 8.09 KB
/
faceformer.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
import torch
import torch.nn as nn
import torch.nn.functional as F
import numpy as np
import copy
import math
from wav2vec import Wav2Vec2Model
# Temporal Bias, inspired by ALiBi: https://github.com/ofirpress/attention_with_linear_biases
def init_biased_mask(n_head, max_seq_len, period):
def get_slopes(n):
def get_slopes_power_of_2(n):
start = (2**(-2**-(math.log2(n)-3)))
ratio = start
return [start*ratio**i for i in range(n)]
if math.log2(n).is_integer():
return get_slopes_power_of_2(n)
else:
closest_power_of_2 = 2**math.floor(math.log2(n))
return get_slopes_power_of_2(closest_power_of_2) + get_slopes(2*closest_power_of_2)[0::2][:n-closest_power_of_2]
slopes = torch.Tensor(get_slopes(n_head))
bias = torch.arange(start=0, end=max_seq_len, step=period).unsqueeze(1).repeat(1,period).view(-1)//(period)
bias = - torch.flip(bias,dims=[0])
alibi = torch.zeros(max_seq_len, max_seq_len)
for i in range(max_seq_len):
alibi[i, :i+1] = bias[-(i+1):]
alibi = slopes.unsqueeze(1).unsqueeze(1) * alibi.unsqueeze(0)
mask = (torch.triu(torch.ones(max_seq_len, max_seq_len)) == 1).transpose(0, 1)
mask = mask.float().masked_fill(mask == 0, float('-inf')).masked_fill(mask == 1, float(0.0))
mask = mask.unsqueeze(0) + alibi
return mask
# Alignment Bias
def enc_dec_mask(device, dataset, T, S):
mask = torch.ones(T, S)
if dataset == "BIWI":
for i in range(T):
mask[i, i*2:i*2+2] = 0
elif dataset == "vocaset":
for i in range(T):
mask[i, i] = 0
return (mask==1).to(device=device)
# Periodic Positional Encoding
class PeriodicPositionalEncoding(nn.Module):
def __init__(self, d_model, dropout=0.1, period=25, max_seq_len=600):
super(PeriodicPositionalEncoding, self).__init__()
self.dropout = nn.Dropout(p=dropout)
pe = torch.zeros(period, d_model)
position = torch.arange(0, period, dtype=torch.float).unsqueeze(1)
div_term = torch.exp(torch.arange(0, d_model, 2).float() * (-math.log(10000.0) / d_model))
pe[:, 0::2] = torch.sin(position * div_term)
pe[:, 1::2] = torch.cos(position * div_term)
pe = pe.unsqueeze(0) # (1, period, d_model)
repeat_num = (max_seq_len//period) + 1
pe = pe.repeat(1, repeat_num, 1)
self.register_buffer('pe', pe)
def forward(self, x):
x = x + self.pe[:, :x.size(1), :]
return self.dropout(x)
class Faceformer(nn.Module):
def __init__(self, args):
super(Faceformer, self).__init__()
"""
audio: (batch_size, raw_wav)
template: (batch_size, V*3)
vertice: (batch_size, seq_len, V*3)
"""
self.dataset = args.dataset
self.audio_encoder = Wav2Vec2Model.from_pretrained("facebook/wav2vec2-base-960h")
# wav2vec 2.0 weights initialization
self.audio_encoder.feature_extractor._freeze_parameters()
self.audio_feature_map = nn.Linear(768, args.feature_dim)
# motion encoder
self.vertice_map = nn.Linear(args.vertice_dim, args.feature_dim)
# periodic positional encoding
self.PPE = PeriodicPositionalEncoding(args.feature_dim, period = args.period)
# temporal bias
self.biased_mask = init_biased_mask(n_head = 4, max_seq_len = 600, period=args.period)
decoder_layer = nn.TransformerDecoderLayer(d_model=args.feature_dim, nhead=4, dim_feedforward=2*args.feature_dim, batch_first=True)
self.transformer_decoder = nn.TransformerDecoder(decoder_layer, num_layers=1)
# motion decoder
self.vertice_map_r = nn.Linear(args.feature_dim, args.vertice_dim)
# style embedding
self.obj_vector = nn.Linear(len(args.train_subjects.split()), args.feature_dim, bias=False)
self.device = args.device
nn.init.constant_(self.vertice_map_r.weight, 0)
nn.init.constant_(self.vertice_map_r.bias, 0)
def forward(self, audio, template, vertice, one_hot, criterion,teacher_forcing=True):
# tgt_mask: :math:`(T, T)`.
# memory_mask: :math:`(T, S)`.
template = template.unsqueeze(1) # (1,1, V*3)
obj_embedding = self.obj_vector(one_hot)#(1, feature_dim)
frame_num = vertice.shape[1]
hidden_states = self.audio_encoder(audio, self.dataset, frame_num=frame_num).last_hidden_state
if self.dataset == "BIWI":
if hidden_states.shape[1]<frame_num*2:
vertice = vertice[:, :hidden_states.shape[1]//2]
frame_num = hidden_states.shape[1]//2
hidden_states = self.audio_feature_map(hidden_states)
if teacher_forcing:
vertice_emb = obj_embedding.unsqueeze(1) # (1,1,feature_dim)
style_emb = vertice_emb
vertice_input = torch.cat((template,vertice[:,:-1]), 1) # shift one position
vertice_input = vertice_input - template
vertice_input = self.vertice_map(vertice_input)
vertice_input = vertice_input + style_emb
vertice_input = self.PPE(vertice_input)
tgt_mask = self.biased_mask[:, :vertice_input.shape[1], :vertice_input.shape[1]].clone().detach().to(device=self.device)
memory_mask = enc_dec_mask(self.device, self.dataset, vertice_input.shape[1], hidden_states.shape[1])
vertice_out = self.transformer_decoder(vertice_input, hidden_states, tgt_mask=tgt_mask, memory_mask=memory_mask)
vertice_out = self.vertice_map_r(vertice_out)
else:
for i in range(frame_num):
if i==0:
vertice_emb = obj_embedding.unsqueeze(1) # (1,1,feature_dim)
style_emb = vertice_emb
vertice_input = self.PPE(style_emb)
else:
vertice_input = self.PPE(vertice_emb)
tgt_mask = self.biased_mask[:, :vertice_input.shape[1], :vertice_input.shape[1]].clone().detach().to(device=self.device)
memory_mask = enc_dec_mask(self.device, self.dataset, vertice_input.shape[1], hidden_states.shape[1])
vertice_out = self.transformer_decoder(vertice_input, hidden_states, tgt_mask=tgt_mask, memory_mask=memory_mask)
vertice_out = self.vertice_map_r(vertice_out)
new_output = self.vertice_map(vertice_out[:,-1,:]).unsqueeze(1)
new_output = new_output + style_emb
vertice_emb = torch.cat((vertice_emb, new_output), 1)
vertice_out = vertice_out + template
loss = criterion(vertice_out, vertice) # (batch, seq_len, V*3)
loss = torch.mean(loss)
return loss
def predict(self, audio, template, one_hot):
template = template.unsqueeze(1) # (1,1, V*3)
obj_embedding = self.obj_vector(one_hot)
hidden_states = self.audio_encoder(audio, self.dataset).last_hidden_state
if self.dataset == "BIWI":
frame_num = hidden_states.shape[1]//2
elif self.dataset == "vocaset":
frame_num = hidden_states.shape[1]
hidden_states = self.audio_feature_map(hidden_states)
for i in range(frame_num):
if i==0:
vertice_emb = obj_embedding.unsqueeze(1) # (1,1,feature_dim)
style_emb = vertice_emb
vertice_input = self.PPE(style_emb)
else:
vertice_input = self.PPE(vertice_emb)
tgt_mask = self.biased_mask[:, :vertice_input.shape[1], :vertice_input.shape[1]].clone().detach().to(device=self.device)
memory_mask = enc_dec_mask(self.device, self.dataset, vertice_input.shape[1], hidden_states.shape[1])
vertice_out = self.transformer_decoder(vertice_input, hidden_states, tgt_mask=tgt_mask, memory_mask=memory_mask)
vertice_out = self.vertice_map_r(vertice_out)
new_output = self.vertice_map(vertice_out[:,-1,:]).unsqueeze(1)
new_output = new_output + style_emb
vertice_emb = torch.cat((vertice_emb, new_output), 1)
vertice_out = vertice_out + template
return vertice_out