-
Notifications
You must be signed in to change notification settings - Fork 1
/
train_gan_face.py
183 lines (156 loc) · 6.54 KB
/
train_gan_face.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
import os
import random
import logging
from PIL import ImageFile, Image
import math
import torch
import torch.nn as nn
import torch.optim as optim
import torch.nn.functional as F
from torch.utils.tensorboard import SummaryWriter
from torch.utils.data import DataLoader
from torchvision import transforms
# from compressai.datasets import ImageFolder
from utils.dataset import FaceImageFolder
from utils.logger import setup_logger
from utils.utils import CustomDataParallel, save_checkpoint
from utils.optimizers import configure_optimizers
from utils.training import train_one_epoch_gan_face
from utils.testing import test_one_epoch_gan_face
from loss.rd_loss import RateDistortionPOELICFaceLoss
from utils.args import train_options
from config.config_5group import model_config
from models.models import ELIC
from models.disc import Discriminator, init_weights
import random
import numpy as np
def setup_seed(seed=3407):
os.environ['PYTHONHASHSEED'] = str(seed)
torch.manual_seed(seed)
torch.cuda.manual_seed(seed)
torch.cuda.manual_seed_all(seed)
np.random.seed(seed)
random.seed(seed)
torch.backends.cudnn.benchmark = False
torch.backends.cudnn.enable = False
return seed
def main():
torch.backends.cudnn.benchmark = True
ImageFile.LOAD_TRUNCATED_IMAGES = True
Image.MAX_IMAGE_PIXELS = None
seed = setup_seed()
args = train_options()
config = model_config()
os.environ['CUDA_VISIBLE_DEVICES'] = ', '.join(str(id) for id in args.gpu_id)
device = "cuda" if args.cuda and torch.cuda.is_available() else "cpu"
if not os.path.exists(os.path.join('./experiments', args.experiment)):
os.makedirs(os.path.join('./experiments', args.experiment))
setup_logger('train', os.path.join('./experiments', args.experiment), 'train_' + args.experiment, level=logging.INFO,
screen=True, tofile=True)
setup_logger('val', os.path.join('./experiments', args.experiment), 'val_' + args.experiment, level=logging.INFO,
screen=True, tofile=True)
logger_train = logging.getLogger('train')
logger_val = logging.getLogger('val')
tb_logger = SummaryWriter(log_dir='./tb_logger/' + args.experiment)
if not os.path.exists(os.path.join('./experiments', args.experiment, 'checkpoints')):
os.makedirs(os.path.join('./experiments', args.experiment, 'checkpoints'))
train_transforms = transforms.Compose(
[transforms.RandomCrop(args.patch_size), transforms.ToTensor()]
)
test_transforms = transforms.Compose(
[transforms.ToTensor()]
)
train_dataset = FaceImageFolder(args.dataset, split="train", transform=train_transforms)
test_dataset = FaceImageFolder(args.dataset, split="kodak", transform=test_transforms)
train_dataloader = DataLoader(
train_dataset,
batch_size=args.batch_size,
num_workers=args.num_workers,
shuffle=True,
pin_memory=(device == "cuda"),
)
test_dataloader = DataLoader(
test_dataset,
batch_size=args.test_batch_size,
num_workers=args.num_workers,
shuffle=False,
pin_memory=(device == "cuda"),
)
net = ELIC(config=config)
net_disc = Discriminator()
if args.cuda and torch.cuda.device_count() > 1:
net = CustomDataParallel(net)
net_disc = CustomDataParallel(net_disc)
net = net.to(device)
net_disc.to(device)
init_weights(net_disc, init_type='normal', init_gain=0.02)
optimizer, aux_optimizer = configure_optimizers(net, args)
# lr_scheduler = optim.lr_scheduler.ReduceLROnPlateau(optimizer, "min")
lr_scheduler = optim.lr_scheduler.MultiStepLR(optimizer, milestones=[80, 100], gamma=0.1)
optimizer_D = torch.optim.Adam(net_disc.parameters(), lr=args.lr_D)
lr_scheduler_D = optim.lr_scheduler.MultiStepLR(optimizer_D, milestones=[80, 100], gamma=0.1)
criterion = RateDistortionPOELICFaceLoss(lmbda=args.lmbda, device=device, gpu_id=args.gpu_id)
if args.checkpoint != None:
checkpoint = torch.load(args.checkpoint)
net.load_state_dict(checkpoint["state_dict"])
optimizer.load_state_dict(checkpoint['optimizer'])
aux_optimizer.load_state_dict(checkpoint['aux_optimizer'])
# lr_scheduler.load_state_dict(checkpoint['lr_scheduler'])
lr_scheduler = optim.lr_scheduler.MultiStepLR(optimizer, milestones=[450,550], gamma=0.1)
lr_scheduler._step_count = checkpoint['lr_scheduler']['_step_count']
lr_scheduler.last_epoch = checkpoint['lr_scheduler']['last_epoch']
# print(lr_scheduler.state_dict())
start_epoch = checkpoint['epoch']
best_loss = checkpoint['loss']
current_step = start_epoch * math.ceil(len(train_dataloader.dataset) / args.batch_size)
checkpoint = None
else:
start_epoch = 0
best_loss = 1e10
current_step = 0
logger_train.info(f"Seed: {seed}")
logger_train.info(args)
# logger_train.info(net)
optimizer.param_groups[0]['lr'] = args.learning_rate
for epoch in range(start_epoch, args.epochs):
logger_train.info(f"Learning rate: {optimizer.param_groups[0]['lr']}")
current_step = train_one_epoch_gan_face(
net,
net_disc,
criterion,
train_dataloader,
optimizer,
aux_optimizer,
optimizer_D,
epoch,
args.clip_max_norm,
logger_train,
tb_logger,
current_step,
config
)
save_dir = os.path.join('./experiments', args.experiment, 'val_images', '%03d' % (epoch + 1))
loss = test_one_epoch_gan_face(epoch, test_dataloader, net, net_disc, criterion, save_dir, logger_val, tb_logger, config)
lr_scheduler.step(loss)
lr_scheduler.step()
lr_scheduler_D.step()
is_best = loss < best_loss
best_loss = min(loss, best_loss)
net.update(force=True)
if args.save :
save_checkpoint(
{
"epoch": epoch + 1,
"state_dict": net.state_dict(),
"loss": loss,
"optimizer": optimizer.state_dict(),
"aux_optimizer": aux_optimizer.state_dict(),
"lr_scheduler": lr_scheduler.state_dict(),
},
is_best,
os.path.join('./experiments', args.experiment, 'checkpoints', "checkpoint_%03d.pth.tar" % (epoch + 1))
)
if is_best:
logger_val.info('best checkpoint saved.')
if __name__ == '__main__':
main()