-
Notifications
You must be signed in to change notification settings - Fork 16
/
train_DINet_clip.py
239 lines (230 loc) · 9.19 KB
/
train_DINet_clip.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
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
import os
import random
import time
import numpy as np
import torch
import torch.nn as nn
import torch.nn.functional as F
import torch.optim as optim
from torch.utils.data import DataLoader
from config.config import DINetTrainingOptions
from dataset.dataset_DINet_clip import DINetDataset
from models.DINet import DINet
from models.Discriminator import Discriminator
from models.Syncnet import SyncNetPerception
from models.VGG19 import Vgg19
from sync_batchnorm import convert_model
from utils.training_utils import GANLoss, get_scheduler, update_learning_rate
if __name__ == "__main__":
"""
clip training code of DINet
in the resolution you want, using clip training code after frame training
"""
# load config
opt = DINetTrainingOptions().parse_args()
random.seed(opt.seed)
np.random.seed(opt.seed)
torch.cuda.manual_seed(opt.seed)
# load training data
train_data = DINetDataset(opt.train_data, opt.augment_num, opt.mouth_region_size)
training_data_loader = DataLoader(
dataset=train_data,
batch_size=opt.batch_size,
shuffle=True,
drop_last=True,
num_workers=2,
)
train_data_length = len(training_data_loader)
# init network
net_g = DINet(opt.source_channel, opt.ref_channel, opt.audio_channel).cuda()
net_dI = Discriminator(
opt.source_channel, opt.D_block_expansion, opt.D_num_blocks, opt.D_max_features
).cuda()
net_dV = Discriminator(
opt.source_channel * 5,
opt.D_block_expansion,
opt.D_num_blocks,
opt.D_max_features,
).cuda()
net_vgg = Vgg19().cuda()
net_lipsync = SyncNetPerception(opt.pretrained_syncnet_path).cuda()
# parallel
net_g = nn.DataParallel(net_g)
net_g = convert_model(net_g)
net_dI = nn.DataParallel(net_dI)
net_dV = nn.DataParallel(net_dV)
net_vgg = nn.DataParallel(net_vgg)
# setup optimizer
optimizer_g = optim.Adam(net_g.parameters(), lr=opt.lr_g)
optimizer_dI = optim.Adam(net_dI.parameters(), lr=opt.lr_dI)
optimizer_dV = optim.Adam(net_dV.parameters(), lr=opt.lr_dI)
# load frame trained DInet weight
print(
"loading frame trained DINet weight from: {}".format(
opt.pretrained_frame_DINet_path
)
)
checkpoint = torch.load(opt.pretrained_frame_DINet_path)
net_g.load_state_dict(checkpoint["state_dict"]["net_g"])
# set criterion
criterionGAN = GANLoss().cuda()
criterionL1 = nn.L1Loss().cuda()
criterionMSE = nn.MSELoss().cuda()
# set scheduler
net_g_scheduler = get_scheduler(optimizer_g, opt.non_decay, opt.decay)
net_dI_scheduler = get_scheduler(optimizer_dI, opt.non_decay, opt.decay)
net_dV_scheduler = get_scheduler(optimizer_dV, opt.non_decay, opt.decay)
# set label of syncnet perception loss
real_tensor = torch.tensor(1.0).cuda()
# start train
for epoch in range(opt.start_epoch, opt.non_decay + opt.decay + 1):
net_g.train()
for iteration, data in enumerate(training_data_loader):
start_time = time.time()
# forward
(
source_clip,
source_clip_mask,
reference_clip,
deep_speech_clip,
deep_speech_full,
) = data
source_clip = (
torch.cat(torch.split(source_clip, 1, dim=1), 0)
.squeeze(1)
.float()
.cuda()
)
source_clip_mask = (
torch.cat(torch.split(source_clip_mask, 1, dim=1), 0)
.squeeze(1)
.float()
.cuda()
)
reference_clip = (
torch.cat(torch.split(reference_clip, 1, dim=1), 0)
.squeeze(1)
.float()
.cuda()
)
deep_speech_clip = (
torch.cat(torch.split(deep_speech_clip, 1, dim=1), 0)
.squeeze(1)
.float()
.cuda()
)
deep_speech_full = deep_speech_full.float().cuda()
fake_out = net_g(source_clip_mask, reference_clip, deep_speech_clip)
fake_out_half = F.avg_pool2d(fake_out, 3, 2, 1, count_include_pad=False)
source_clip_half = F.interpolate(
source_clip, scale_factor=0.5, mode="bilinear"
)
# (1) Update DI network
optimizer_dI.zero_grad()
_, pred_fake_dI = net_dI(fake_out)
loss_dI_fake = criterionGAN(pred_fake_dI, False)
_, pred_real_dI = net_dI(source_clip)
loss_dI_real = criterionGAN(pred_real_dI, True)
# Combined DI loss
loss_dI = (loss_dI_fake + loss_dI_real) * 0.5
loss_dI.backward(retain_graph=True)
optimizer_dI.step()
# (2) Update DV network
optimizer_dV.zero_grad()
condition_fake_dV = torch.cat(
torch.split(fake_out, opt.batch_size, dim=0), 1
)
_, pred_fake_dV = net_dV(condition_fake_dV)
loss_dV_fake = criterionGAN(pred_fake_dV, False)
condition_real_dV = torch.cat(
torch.split(source_clip, opt.batch_size, dim=0), 1
)
_, pred_real_dV = net_dV(condition_real_dV)
loss_dV_real = criterionGAN(pred_real_dV, True)
# Combined DV loss
loss_dV = (loss_dV_fake + loss_dV_real) * 0.5
loss_dV.backward(retain_graph=True)
optimizer_dV.step()
# (2) Update DINet
_, pred_fake_dI = net_dI(fake_out)
_, pred_fake_dV = net_dV(condition_fake_dV)
optimizer_g.zero_grad()
# compute perception loss
perception_real = net_vgg(source_clip)
perception_fake = net_vgg(fake_out)
perception_real_half = net_vgg(source_clip_half)
perception_fake_half = net_vgg(fake_out_half)
loss_g_perception = 0
for i in range(len(perception_real)):
loss_g_perception += criterionL1(perception_fake[i], perception_real[i])
loss_g_perception += criterionL1(
perception_fake_half[i], perception_real_half[i]
)
loss_g_perception = (
loss_g_perception / (len(perception_real) * 2)
) * opt.lamb_perception
# # gan dI loss
loss_g_dI = criterionGAN(pred_fake_dI, True)
# # gan dV loss
loss_g_dV = criterionGAN(pred_fake_dV, True)
# sync perception loss
fake_out_clip = torch.cat(torch.split(fake_out, opt.batch_size, dim=0), 1)
fake_out_clip_mouth = fake_out_clip[
:,
:,
train_data.radius : train_data.radius + train_data.mouth_region_size,
train_data.radius_1_4 : train_data.radius_1_4
+ train_data.mouth_region_size,
]
sync_score = net_lipsync(fake_out_clip_mouth, deep_speech_full)
loss_sync = (
criterionMSE(sync_score, real_tensor.expand_as(sync_score))
* opt.lamb_syncnet_perception
)
# combine all losses
loss_g = loss_g_perception + loss_g_dI + loss_g_dV + loss_sync
loss_g.backward()
optimizer_g.step()
end_time = time.time() # End time of the epoch
elapsed_time = end_time - start_time
print(
"===> Epoch[{}]({}/{}): Loss_DI: {:.4f} Loss_GI: {:.4f} Loss_DV: {:.4f} Loss_GV: {:.4f} Loss_perception: {:.4f} Loss_sync: {:.4f} lr_g = {:.7f} elapsed_time: {:.4f}".format(
epoch,
iteration,
len(training_data_loader),
float(loss_dI),
float(loss_g_dI),
float(loss_dV),
float(loss_g_dV),
float(loss_g_perception),
float(loss_sync),
optimizer_g.param_groups[0]["lr"],
),
len(training_data_loader) / elapsed_time,
)
# iterations_per_sec = len(training_data_loader) / elapsed_time
update_learning_rate(net_g_scheduler, optimizer_g)
update_learning_rate(net_dI_scheduler, optimizer_dI)
update_learning_rate(net_dV_scheduler, optimizer_dV)
# checkpoint
if epoch % opt.checkpoint == 0:
if not os.path.exists(opt.result_path):
os.mkdir(opt.result_path)
model_out_path = os.path.join(
opt.result_path, "netG_model_epoch_{}.pth".format(epoch)
)
states = {
"epoch": epoch + 1,
"state_dict": {
"net_g": net_g.state_dict(),
"net_dI": net_dI.state_dict(),
"net_dV": net_dV.state_dict(),
},
"optimizer": {
"net_g": optimizer_g.state_dict(),
"net_dI": optimizer_dI.state_dict(),
"net_dV": optimizer_dV.state_dict(),
},
}
torch.save(states, model_out_path)
print("Checkpoint saved to {}".format(epoch))