forked from icon-lab/ResViT
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathtrain.py
167 lines (151 loc) · 7.11 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
import time
from options.train_options import TrainOptions
from data import CreateDataLoader
from models import create_model
from util.visualizer import Visualizer
import numpy as np, h5py
from skimage.metrics import peak_signal_noise_ratio as psnr
from skimage.metrics import structural_similarity as ssim
#from skimage.measure import compare_psnr as psnr
import resource
rlimit = resource.getrlimit(resource.RLIMIT_NOFILE)
resource.setrlimit(resource.RLIMIT_NOFILE, (4096 * 8, rlimit[1]))
import os
from tqdm import tqdm
import wandb
def print_log(logger,message):
print(message, flush=True)
if logger:
logger.write(str(message) + '\n')
if __name__ == '__main__':
opt = TrainOptions().parse()
#Training data
opt.phase='train'
data_loader = CreateDataLoader(opt)
#dataset = data_loader.load_data()
dataset_size = len(data_loader.dataset)
print('#Training images = %d' % dataset_size)
##logger ##
save_dir = os.path.join(opt.checkpoints_dir, opt.name)
logger = open(os.path.join(save_dir, 'log.txt'), 'w+')
print_log(logger,opt.name)
logger.close()
#9ba56f9cf26aa9a552c1fc73fd318072f2f85a24
if not opt.no_wandb:
wandb.login()
wandb.init(project="PseudoPET-AD-ResViT",entity="dperlo_lih",name=opt.name,config=vars(opt))
#validation data
opt.phase='val'
data_loader_val = CreateDataLoader(opt)
#dataset_val = data_loader_val.load_data()
dataset_size_val = len(data_loader_val.dataset)
print('#Validation images = %d' % dataset_size_val)
#if opt.model=='cycle_gan':
# L1_avg=np.zeros([2,opt.niter + opt.niter_decay,len(data_loader_val)])
# psnr_avg=np.zeros([2,opt.niter + opt.niter_decay,len(data_loader_val)])
#else:
L1_avg=np.zeros([opt.niter + opt.niter_decay,len(data_loader_val)])
psnr_avg=np.zeros([opt.niter + opt.niter_decay,len(data_loader_val)])
ssim_avg=np.zeros([opt.niter + opt.niter_decay,len(data_loader_val)])
model = create_model(opt)
visualizer = Visualizer(opt)
total_steps = 0
for epoch in range(opt.epoch_count, opt.niter + opt.niter_decay + 1):
epoch_start_time = time.time()
iter_data_time = time.time()
epoch_iter = 0
#Training step
opt.phase='train'
for i, data in tqdm(enumerate(data_loader)):
iter_start_time = time.time()
if total_steps % opt.print_freq == 0:
t_data = iter_start_time - iter_data_time
visualizer.reset()
total_steps += opt.batchSize
epoch_iter += opt.batchSize
model.set_input(data)
model.optimize_parameters()
if epoch_iter % opt.display_freq == 0 and opt.print_res:
save_result = total_steps % opt.update_html_freq == 0
if opt.dataset_mode=='aligned_mat':
temp_visuals=model.get_current_visuals()
print(temp_visuals.shape)
visualizer.display_current_results(temp_visuals, epoch, save_result, epoch_iter)
elif opt.dataset_mode=='unaligned_mat':
temp_visuals=model.get_current_visuals()
temp_visuals['real_A']=temp_visuals['real_A'][:,:,0:3]
temp_visuals['real_B']=temp_visuals['real_B'][:,:,0:3]
temp_visuals['fake_A']=temp_visuals['fake_A'][:,:,0:3]
temp_visuals['fake_B']=temp_visuals['fake_B'][:,:,0:3]
temp_visuals['rec_A']=temp_visuals['rec_A'][:,:,0:3]
temp_visuals['rec_B']=temp_visuals['rec_B'][:,:,0:3]
if opt.lambda_identity>0:
temp_visuals['idt_A']=temp_visuals['idt_A'][:,:,0:3]
temp_visuals['idt_B']=temp_visuals['idt_B'][:,:,0:3]
visualizer.display_current_results(temp_visuals, epoch, save_result, epoch_iter)
else:
temp_visuals=model.get_current_visuals()
visualizer.display_current_results(temp_visuals, epoch, save_result, epoch_iter)
if epoch_iter % opt.print_freq == 0:
errors = model.get_current_errors()
t = (time.time() - iter_start_time) / opt.batchSize
visualizer.print_current_errors(epoch, epoch_iter, errors, t, t_data)
if not opt.no_wandb:
wandb.log(errors)
if opt.display_id > 0:
visualizer.plot_current_errors(epoch, float(epoch_iter) / dataset_size, opt, errors)
if epoch_iter % opt.save_latest_freq == 0:
print('saving the latest model (epoch %d, total_steps %d)' %
(epoch, total_steps))
model.save('latest')
iter_data_time = time.time()
#Validaiton step
if epoch % opt.val_step == 0:
logger = open(os.path.join(save_dir, 'log.txt'), 'a')
print(opt.dataset_mode)
opt.phase='val'
for i, data_val in enumerate(data_loader_val):
#
model.set_input(data_val)
#
model.test()
#
fake_B = model.fake_B.view(-1,opt.fineSize,opt.fineSize)
fake_im= fake_B.cpu().data.numpy()
#
real_B = model.real_B.view(-1,opt.fineSize,opt.fineSize)
real_im= real_B.cpu().data.numpy()
#
real_im=real_im*0.5+0.5
fake_im=fake_im*0.5+0.5
#if real_im.max() <= 0:
# continue
L1_avg[epoch-1,i]=abs(fake_im-real_im).mean()
psnr_avg[epoch-1,i]=psnr(fake_im,real_im, data_range=1)#psnr(fake_im/fake_im.max(),real_im/real_im.max())
ssim_avg[epoch-1,i]=ssim(fake_im,real_im, channel_axis=0, data_range=1) #ssim(fake_im/fake_im.max(),real_im/real_im.max(), channel_axis=opt.batchSize)
#
#
l1_avg_loss = np.mean(L1_avg[epoch-1])
#
mean_psnr = np.mean(psnr_avg[epoch-1])
mean_ssim = np.mean(ssim_avg[epoch-1])
#
std_psnr = np.std(psnr_avg[epoch-1])
#
print_log(logger,'Epoch %3d l1_avg_loss: %.5f mean_psnr: %.3f std_psnr:%.3f mean_ssim: %.3f' % \
(epoch, l1_avg_loss, mean_psnr,std_psnr,mean_ssim))
#
print_log(logger,'')
logger.close()
if not opt.no_wandb:
wandb.log({"l1_avg_loss":l1_avg_loss,"mean_psnr":mean_psnr,"std_psnr":std_psnr,"mean_ssim":mean_ssim})
#
if epoch % opt.save_epoch_freq == 0:
print('saving the model at the end of epoch %d, iters %d' %(epoch, total_steps))
#
model.save('latest')
#
model.save(epoch)
print('End of epoch %d / %d \t Time Taken: %d sec' %
(epoch, opt.niter + opt.niter_decay, time.time() - epoch_start_time))
model.update_learning_rate()