-
Notifications
You must be signed in to change notification settings - Fork 6
/
train.py
171 lines (130 loc) · 6.85 KB
/
train.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
from __future__ import print_function
import os
from math import log10
from collections import OrderedDict
import torchvision.utils as vutils
import torch
import torch.nn as nn
import torch.nn.functional as f
from torch.utils.data import DataLoader
from dataset import DatasetFromFolder
import torch.backends.cudnn as cudnn
from PIL import Image
import numpy as np
from pix2pix_pro import Pix2PixPro
import util.util as util
from util.visualizer import Visualizer
import time
from options.train_options import TrainOptions, EPOCH_MAPPING, MINIBATCH_MAPPING
opt = TrainOptions().parse()
import random
random.seed(2143155159)
if __name__ == '__main__':
cudnn.benchmark = True
print("GAN_MODE: {}".format(opt.gan_mode))
visualizer = Visualizer(opt)
torch.manual_seed(opt.seed)
if torch.cuda.is_available():
torch.cuda.manual_seed(opt.seed)
## creating checkpoint dir
if not os.path.exists("result"):
os.mkdir("result")
if not os.path.exists(os.path.join("result", opt.project_name)):
os.mkdir(os.path.join("result", opt.project_name))
print('===> Loading datasets')
root_path = opt.dataset_root
pix2pix_pro = Pix2PixPro()
pix2pix_pro.initialize(opt)
pix2pix_pro = nn.DataParallel(pix2pix_pro, device_ids=opt.gpu_ids)
pix2pix_pro.cuda()
current_resolution = opt.start_resolution
print("===> start epoch:{} start resolution:{}".format(opt.start_epoch, opt.start_resolution))
for epoch in range(opt.start_epoch, opt.niter + opt.niter_decay + 1):
random.seed(2143155159 * (epoch+1))
net_status = "stable"
if epoch == 1 and current_resolution == 256:
net_status = "stable"
elif epoch > EPOCH_MAPPING[current_resolution]:
if current_resolution != 256:
current_resolution *= 2
net_status = "fadein"
print("==> fadeIn! current_resolution:{}".format(current_resolution))
psnr_cache = 0
else:
net_status = "stable"
pix2pix_pro.module.set_config(current_resolution, net_status, 1.0)
batch_size = MINIBATCH_MAPPING[current_resolution] * opt.batch_rate
if net_status == "fadein":
batch_size = int(batch_size/2)
if(batch_size < 1):
batch_size = 1
train_set = DatasetFromFolder(root_path + opt.dataset + "/train", opt.direction, current_resolution, is_train=True)
training_data_loader = DataLoader(dataset=train_set, num_workers=opt.threads, batch_size=batch_size,
shuffle=True)
test_set = DatasetFromFolder(root_path + opt.dataset + "/test", opt.direction, current_resolution, is_train=False)
testing_data_loader = DataLoader(dataset=test_set, num_workers=2, batch_size=batch_size,
shuffle=False)
# train
for iteration, batch in enumerate(training_data_loader, 1):
if iteration % 10 == 0:
iter_start_time = time.time()
if net_status == "fadein":
net_alpha = 1.0 - (iteration + 1) / len(train_set)
pix2pix_pro.module.set_config(current_resolution, "fadein", net_alpha)
# forward
real_a, real_b, real_256_a, other_a = f.interpolate(batch[2], current_resolution).cuda(), batch[1].cuda(), batch[2].cuda(), batch[6].cuda()
losses, generated = pix2pix_pro(real_256_a, real_a, real_b, other_a)
losses = {k: v.mean() if not isinstance(v, int) else v for k, v in losses.items()}
loss_d = ( losses["D_fake"] + losses["D_real"] + losses["D_other"] ) / 3
loss_d_tf = ( losses["D_tf_fake"] + losses["D_tf_real"] ) / 2
loss_g = losses["G_GAN"] + losses["G_GAN_tf"] + losses["G_FM"] + losses.get("G_FM_tf", 0) + losses.get("G_Input", 0) + losses["G_L1"]
pix2pix_pro.module.optimizer_g.zero_grad()
loss_g.backward()
pix2pix_pro.module.optimizer_g.step()
pix2pix_pro.module.optimizer_d.zero_grad()
loss_d.backward()
pix2pix_pro.module.optimizer_d.step()
pix2pix_pro.module.optimizer_d_tf.zero_grad()
loss_d_tf.backward()
pix2pix_pro.module.optimizer_d_tf.step()
############## Display results and errors ##########
### print out errors
if iteration % 10 == 0:
errors = {k: v.data.item() if not isinstance(v, int) else v for k, v in losses.items()}
t = (time.time() - iter_start_time) / 10
visualizer.print_current_errors(epoch, iteration, errors, t)
total_steps = epoch * len(training_data_loader) + iteration
visualizer.plot_current_errors(errors, total_steps)
visuals = OrderedDict([('input_label', util.tensor2im(real_a.data[0])),
('synthesized_image', util.tensor2im(generated.data[0])),
('real_image', util.tensor2im(real_b.data[0]))])
visualizer.display_current_results(visuals, epoch, total_steps)
pix2pix_pro.module.update_learning_rate()
pix2pix_pro.module.save("latest")
# test
avg_psnr = 0
for iteration,batch in enumerate(testing_data_loader):
real_a, real_b, real_256_a, real_256_b = \
batch[0].cuda(), batch[1].cuda(), batch[2].cuda(), batch[3].cuda()#, batch[4].to(device)#, batch[5].to(device)#, batch[6].to(device)
prediction = pix2pix_pro.module.inference(real_256_a.detach())
mse = np.mean((prediction.cpu().numpy() - real_b.cpu().numpy()) ** 2)
psnr = 10 * log10(1 / mse.item())
avg_psnr += psnr
vutils.save_image(torch.cat((real_256_a, f.interpolate(prediction.float(), 256), real_256_b), 3),
"{}/{}/test_result_epoch_{}_iter_{}".format(opt.checkpoints_dir, opt.project_name, str(epoch), str(iteration)) + ".jpg",
nrow=4, normalize=True, padding=0)
del prediction, real_a, real_b, real_256_a, real_256_b
result_img = Image.new('RGB', (0, 0), (0, 0, 0))
for p in range(99):
try:
img_name = "{}/{}/test_result_epoch_{}_iter_{}".format(opt.checkpoints_dir, opt.project_name, str(epoch), str(p)) + ".jpg"
im = Image.open(img_name)
result_img = util.get_concat_v_blank(result_img, im)
os.remove(img_name)
except:
break
result_img.save("{}/{}/test_result_epoch_{}".format(opt.checkpoints_dir, opt.project_name, str(epoch)) + ".jpg")
avg_psnr = avg_psnr / len(testing_data_loader)
print("===> Avg. PSNR: {:.4f} dB".format(avg_psnr))
if epoch % 10 == 0 and epoch != 0:
pix2pix_pro.module.save(epoch)