-
Notifications
You must be signed in to change notification settings - Fork 20
/
run.py
257 lines (215 loc) · 12.3 KB
/
run.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
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
import shutil
import matplotlib
matplotlib.use('Agg')
import os, sys
import yaml
from argparse import ArgumentParser
from time import gmtime, strftime
from shutil import copy
from frames_dataset import FramesDataset
# from frames_dataset_zoom import FramesDataset as FramesDataset_zoom
import pdb
import modules.generator as G
from modules.discriminator import MultiScaleDiscriminator
from modules.keypoint_detector import KPDetector
import modules.keypoint_detector as KPD
import torch.distributed as dist
from torch.nn.parallel import DistributedDataParallel as DDP
import torch
from torch.utils.tensorboard import SummaryWriter
import train
from reconstruction import reconstruction
from animate import animate
import random
import numpy as np
from collections import OrderedDict
def seed_torch(seed=1029):
random.seed(seed)
os.environ['PYTHONHASHSEED'] = str(seed) # 为了禁止hash随机化，使得实验可复现
np.random.seed(seed)
torch.manual_seed(seed)
torch.cuda.manual_seed(seed)
torch.cuda.manual_seed_all(seed) # if you are using multi-GPU.
torch.backends.cudnn.benchmark = True
torch.backends.cudnn.deterministic = True
seed_torch()
def mb_load(ckp, model,flag):
model.mbUnit.mb.data = ckp['generator']['module.mbUnit.mb']
if flag == 'finetune':
model.mbUnit.mb.requires_grad = True
if flag == 'fixed':
model.mbUnit.mb.requires_grad = False
return model
if __name__ == "__main__":
if sys.version_info[0] < 3:
raise Exception("You must use Python 3 or higher. Recommended version is Python 3.7")
parser = ArgumentParser()
parser.add_argument("--config", required=True, help="path to config")
parser.add_argument("--mode", default="train", choices=["train", "reconstruction", "animate","train_avd"])
parser.add_argument("--log_dir", default='log', help="path to log into")
parser.add_argument("--checkpoint", default=None, help="path to checkpoint to restore")
parser.add_argument("--device_ids", default="0", type=lambda x: list(map(int, x.split(','))),
help="Names of the devices comma separated.")
parser.add_argument("--verbose", dest="verbose", action="store_true", help="Print model architecture")
parser.add_argument("--local_rank", type=int)
parser.add_argument("--use_depth",action='store_true',help='depth mode')
parser.add_argument("--rgbd",action='store_true',help='depth mode')
# alter model
parser.add_argument("--generator",required=True,help='depth mode')
parser.add_argument("--kp_detector",default='KPDetector',type=str,help='depth mode')
parser.add_argument("--GFM",default='GeneratorFullModel')
parser.add_argument("--batchsize",type=int, default=-1,help='depth mode')
parser.add_argument("--kp_num",type=int, default=-1,help='depth mode')
parser.add_argument("--kp_distance",type=int, default=0,help='depth mode')
parser.add_argument("--depth_constraint",type=int, default=0,help='depth mode')
parser.add_argument("--app_mask",type=float, default=0, help='depth mode')
parser.add_argument("--mb_consistent",type=float, default=0, help='depth mode')
parser.add_argument("--feat_consistent",type=float, default=0, help='depth mode')
parser.add_argument("--bi_feat_consistent",type=float, default=0, help='depth mode')
parser.add_argument("--l1",type=float, default=0, help='depth mode')
parser.add_argument("--image_ffl",type=float, default=0, help='depth mode')
parser.add_argument("--feat_ffl",type=float, default=0, help='depth mode')
parser.add_argument("--warp_loss",type=float, default=0, help='depth mode')
parser.add_argument("--mb_ffl",type=float, default=0, help='depth mode')
parser.add_argument("--occlusion_smooth",type=float, default=0, help='depth mode')
parser.add_argument("--hierachy_constraint",type=float, default=0, help='depth mode')
parser.add_argument("--equivariance_keypoint",type=float, default=0, help='depth mode')
parser.add_argument("--kl_mb",type=float, default=0, help='depth mode')
parser.add_argument("--kl_feat",type=float, default=0, help='depth mode')
parser.add_argument("--vq_commit",type=float, default=0, help='depth mode')
parser.add_argument("--attn_regular",type=float, default=0, help='depth mode')
parser.add_argument("--qv_style_similar",type=float, default=0, help='depth mode')
parser.add_argument("--feat_gap",type=float, default=0, help='depth mode')
parser.add_argument("--sample_feat_consistent",type=float, default=0, help='depth mode')
parser.add_argument("--reconstruction",type=float, default=0, help='depth mode')
parser.add_argument("--kp_prior",type=float, default=0, help='depth mode')
parser.add_argument("--identity",type=float, default=0, help='depth mode')
parser.add_argument("--FDIT",type=float, default=0, help='depth mode')
parser.add_argument("--mbunit",type=str, default='', help='depth mode')
parser.add_argument("--clip_grad",type=float, default=10, help='depth mode')
parser.add_argument("--mb_pretrained",action='store_true')
parser.add_argument("--mb_finetune",action='store_true')
parser.add_argument("--single_optim",action='store_true')
parser.add_argument("--styleGAN",action='store_true')
parser.add_argument("--sft_cross",action='store_true')
parser.add_argument("--multi_scale",action='store_true')
parser.add_argument("--mb_channel",type=int, default=512, help='depth mode')
parser.add_argument("--mb_spatial",type=int, default=32, help='depth mode')
parser.add_argument("--generator_gan",type=float, default=1, help='depth mode')
parser.add_argument("--memsize",type=int, default=8, help='depth mode')
parser.add_argument("--linear_grow_mb_weight",action='store_true')
parser.add_argument("--depth_path",type=str,help='the path of depth weight')
parser.add_argument("--no_jacobain",action='store_true')
parser.add_argument("--no_disc_use_kp",action='store_true')
parser.add_argument("--name",type=str)
parser.set_defaults(verbose=False)
opt = parser.parse_args()
with open(opt.config) as f:
config = yaml.load(f)
if opt.checkpoint is not None:
if opt.mode != "train_avd":
log_dir = os.path.join(*os.path.split(opt.checkpoint)[:-1])
else:
log_dir = os.path.join(opt.log_dir, opt.name)
else:
log_dir = os.path.join(opt.log_dir, opt.name)
if not os.path.exists(log_dir):
os.makedirs(log_dir)
print("Training...")
dist.init_process_group(backend='nccl', init_method='env://')
torch.cuda.set_device(opt.local_rank)
device=torch.device("cuda",opt.local_rank)
config['train_params']['loss_weights']['equivariance_keypoint'] = opt.equivariance_keypoint
config['train_params']['loss_weights']['depth_constraint'] = opt.depth_constraint
config['train_params']['loss_weights']['kp_distance'] = opt.kp_distance
config['train_params']['loss_weights']['app_mask'] = opt.app_mask
config['train_params']['loss_weights']['mb_consistent'] = opt.mb_consistent
config['train_params']['loss_weights']['occlusion_smooth'] = opt.occlusion_smooth
config['train_params']['loss_weights']['hierachy_constraint'] = opt.hierachy_constraint
config['train_params']['loss_weights']['kl_mb'] = opt.kl_mb
config['train_params']['loss_weights']['kl_feat'] = opt.kl_feat
config['train_params']['loss_weights']['feat_consistent'] = opt.feat_consistent
config['train_params']['loss_weights']['vq_commit'] = opt.vq_commit
config['train_params']['loss_weights']['attn_regular'] = opt.attn_regular
config['train_params']['loss_weights']['bi_feat_consistent'] = opt.bi_feat_consistent
config['train_params']['loss_weights']['qv_style_similar'] = opt.qv_style_similar
config['train_params']['loss_weights']['feat_gap'] = opt.feat_gap
config['train_params']['loss_weights']['generator_gan'] = opt.generator_gan
config['train_params']['loss_weights']['sample_feat_consistent'] = opt.sample_feat_consistent
config['train_params']['loss_weights']['l1'] = opt.l1
config['train_params']['loss_weights']['warp_loss'] = opt.warp_loss
config['train_params']['loss_weights']['reconstruction'] = opt.reconstruction
config['train_params']['loss_weights']['kp_prior'] = opt.kp_prior
config['train_params']['loss_weights']['image_ffl'] = opt.image_ffl
config['train_params']['loss_weights']['feat_ffl'] = opt.feat_ffl
config['train_params']['loss_weights']['mb_ffl'] = opt.mb_ffl
config['train_params']['loss_weights']['identity'] = opt.identity
config['train_params']['loss_weights']['FDIT'] = opt.FDIT
config['model_params']['generator_params']['memsize'] = opt.memsize
if opt.no_jacobain:
config['model_params']['common_params']['estimate_jacobian'] = False
config['train_params']['loss_weights']['equivariance_jacobian'] = 0
if opt.batchsize != -1:
config['train_params']['batch_size'] = opt.batchsize
if opt.kp_num != -1:
config['model_params']['common_params']['num_kp'] = opt.kp_num
if opt.no_disc_use_kp:
config['model_params']['discriminator_params']['use_kp'] = False
# create generator
generator = getattr(G, opt.generator)(**config['model_params']['generator_params'],**config['model_params']['common_params'],**{'mbunit':opt.mbunit,'mb_spatial':opt.mb_spatial,'mb_channel':opt.mb_channel})
if opt.mb_pretrained:
checkpoint_path = 'log/Unet_Reconstruction_no_adv/00000099-checkpoint.pth.tar'
ckp = torch.load(checkpoint_path, map_location=torch.device('cpu'))
generator = mb_load(ckp,generator,flag='fixed')
if opt.mb_finetune:
checkpoint_path = 'log/Unet_Reconstruction_no_adv/00000099-checkpoint.pth.tar'
ckp = torch.load(checkpoint_path, map_location=torch.device('cpu'))
generator = mb_load(ckp,generator,flag='finetune')
generator.to(device)
if opt.verbose:
print(generator)
generator= torch.nn.SyncBatchNorm.convert_sync_batchnorm(generator)
# create discriminator
discriminator = MultiScaleDiscriminator(**config['model_params']['discriminator_params'],
**config['model_params']['common_params'])
discriminator.to(device)
if opt.verbose:
print(discriminator)
discriminator= torch.nn.SyncBatchNorm.convert_sync_batchnorm(discriminator)
kp_detector = getattr(KPD, opt.kp_detector)(**config['model_params']['kp_detector_params'],
**config['model_params']['common_params'])
kp_detector.to(device)
if opt.verbose:
print(kp_detector) >> 'd.txt'
kp_detector= torch.nn.SyncBatchNorm.convert_sync_batchnorm(kp_detector)
#Create the record of network arch
# with open(os.path.join(log_dir,'network.txt'),"w") as file:
# file.write(kp_detector)
# file.write(discriminator)
# file.write(generator)
if torch.cuda.device_count() == 1:
kp_detector = DDP(kp_detector,device_ids=[opt.local_rank],broadcast_buffers=False)
discriminator = DDP(discriminator,device_ids=[opt.local_rank],broadcast_buffers=False)
generator = DDP(generator,device_ids=[opt.local_rank],broadcast_buffers=False)
else:
kp_detector = DDP(kp_detector,device_ids=[opt.local_rank])
discriminator = DDP(discriminator,device_ids=[opt.local_rank])
generator = DDP(generator,device_ids=[opt.local_rank])
dataset = FramesDataset(is_train=(opt.mode == 'train' or opt.mode == 'train_avd'), **config['dataset_params'])
dataset.__getitem__(0)
if not os.path.exists(log_dir):
os.makedirs(log_dir)
if not os.path.exists(os.path.join(log_dir, os.path.basename(opt.config))):
copy(opt.config, log_dir)
if not os.path.exists(os.path.join(log_dir,'log')):
os.makedirs(os.path.join(log_dir,'log'))
writer = SummaryWriter(os.path.join(log_dir,'log'))
# print(config)
if opt.mode == 'train':
train.train(config, generator, discriminator, kp_detector, opt.checkpoint, log_dir, dataset, opt.local_rank,device,opt,writer)
elif opt.mode == 'reconstruction':
print("Reconstruction...")
reconstruction(config, generator, kp_detector, opt.checkpoint, log_dir, dataset)
elif opt.mode == 'animate':
print("Animate...")
animate(config, generator, kp_detector, opt.checkpoint, log_dir, dataset,opt)