-
Notifications
You must be signed in to change notification settings - Fork 8
/
train.py
205 lines (191 loc) · 10.7 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
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
import os.path
import sys
import math
import argparse
import time
import random
from collections import OrderedDict,deque
import matplotlib
matplotlib.use('Agg')
import torch
import numpy as np
import options.options as option
from utils import util
from data import create_dataloader, create_dataset
from models import create_model
from utils.logger import Logger, PrintLogger
from datetime import datetime
IGNORED_KEYS_LIST = ['l_d_real','l_d_fake','D_real','D_fake','psnr_val','LR_decrease','Correctly_distinguished','l_g_range','D_loss_STD']#,'l_g_pix'
def main():
# options
parser = argparse.ArgumentParser()
parser.add_argument('-opt', type=str, required=True, help='Path to option JSON file.')
parser.add_argument('-single_GPU', action='store_true',help='Utilize only one GPU')
if parser.parse_args().single_GPU:
available_GPUs = util.Assign_GPU()
else:
available_GPUs = util.Assign_GPU(max_GPUs=None)
opt = option.parse(parser.parse_args().opt, is_train=True,batch_size_multiplier=len(available_GPUs))
if not opt['train']['resume']:
util.mkdir_and_rename(opt['path']['experiments_root']) # Modify experiment name if exists
util.mkdirs((path for key, path in opt['path'].items() if not key == 'experiments_root' and \
not key == 'pretrained_model_G' and not key == 'pretrained_model_D'))
option.save(opt)
opt = option.dict_to_nonedict(opt) # Convert to NoneDict, which return None for missing key.
# print to file and std_out simultaneously
sys.stdout = PrintLogger(opt['path']['log'])
# random seed
seed = opt['train']['manual_seed']
if seed is None:
seed = random.randint(1, 10000)
print("Random Seed: ", seed)
random.seed(seed)
torch.manual_seed(seed)
# create train and val dataloader
for phase, dataset_opt in opt['datasets'].items():
if phase == 'train':
max_accumulation_steps = max([opt['train']['grad_accumulation_steps_G'], opt['train']['grad_accumulation_steps_D']])
train_set = create_dataset(dataset_opt)
train_size = int(math.ceil(len(train_set) / dataset_opt['batch_size']))
print('Number of train images: {:,d}, iters: {:,d}'.format(len(train_set), train_size))
total_iters = int(opt['train']['niter']*max_accumulation_steps)#-current_step
total_epoches = int(math.ceil(total_iters / train_size))
print('Total epoches needed: {:d} for iters {:,d}'.format(total_epoches, total_iters))
train_loader = create_dataloader(train_set, dataset_opt)
elif phase == 'val':
val_dataset_opt = dataset_opt
val_set = create_dataset(dataset_opt)
val_loader = create_dataloader(val_set, dataset_opt)
print('Number of val images in [{:s}]: {:d}'.format(dataset_opt['name'], len(val_set)))
else:
raise NotImplementedError('Phase [{:s}] is not recognized.'.format(phase))
assert train_loader is not None
# Create model
if max_accumulation_steps!=1:
model = create_model(opt,max_accumulation_steps)
else:
model = create_model(opt)
# create logger
logger = Logger(opt)
# Save validation set results as image collage:
SAVE_IMAGE_COLLAGE = True
per_image_saved_patch = min([min(im['HR'].shape[1:]) for im in val_loader.dataset])-2
num_val_images = len(val_loader.dataset)
val_images_collage_rows = int(np.floor(np.sqrt(num_val_images)))
while val_images_collage_rows>1:
if np.round(num_val_images/val_images_collage_rows)==num_val_images/val_images_collage_rows:
break
val_images_collage_rows -= 1
start_time = time.time()
min_accumulation_steps = min([opt['train']['grad_accumulation_steps_G'],opt['train']['grad_accumulation_steps_D']])
save_GT_HR = True
lr_too_low = False
print('---------- Start training -------------')
last_saving_time = time.time()
recently_saved_models = deque(maxlen=4)
for epoch in range(int(math.floor(model.step / train_size)),total_epoches):
for i, train_data in enumerate(train_loader):
model.gradient_step_num = model.step // max_accumulation_steps
not_within_batch = model.step % max_accumulation_steps == (max_accumulation_steps - 1)
saving_step = (model.gradient_step_num==0 or (time.time()-last_saving_time)>60*opt['logger']['save_checkpoint_freq']) and not_within_batch
if saving_step:
last_saving_time = time.time()
# save models
if lr_too_low or saving_step:
recently_saved_models.append(model.save(model.gradient_step_num))
model.save_log()
if len(recently_saved_models)>3:
model_2_delete = recently_saved_models.popleft()
os.remove(model_2_delete)
if model.D_exists:
os.remove(model_2_delete.replace('_G.','_D.'))
print('{}: Saving the model before iter {:d}.'.format(datetime.now().strftime('%H:%M:%S'),model.gradient_step_num))
if lr_too_low:
break
if model.step > total_iters:
break
# training
model.feed_data(train_data)
model.optimize_parameters()
if not model.D_exists:#Avoid using the naive MultiLR scheduler when using adversarial loss
for scheduler in model.schedulers:
scheduler.step(model.gradient_step_num)
time_elapsed = time.time() - start_time
if not_within_batch: start_time = time.time()
# log
if model.gradient_step_num % opt['logger']['print_freq'] == 0 and not_within_batch:
logs = model.get_current_log()
print_rlt = OrderedDict()
print_rlt['model'] = opt['model']
print_rlt['epoch'] = epoch
print_rlt['iters'] = model.gradient_step_num
print_rlt['time'] = time_elapsed
for k, v in logs.items():
print_rlt[k] = v
print_rlt['lr'] = model.get_current_learning_rate()
logger.print_format_results('train', print_rlt,keys_ignore_list=IGNORED_KEYS_LIST)
model.display_log_figure()
# validation
if not_within_batch and (model.gradient_step_num) % opt['train']['val_freq'] == 0: # and model.gradient_step_num>=opt['train']['D_init_iters']:
print_rlt = OrderedDict()
if model.generator_changed:
print('---------- validation -------------')
start_time = time.time()
if False and SAVE_IMAGE_COLLAGE and model.gradient_step_num%opt['train']['val_save_freq'] == 0: #Saving training images:
GT_image_collage = []
cur_train_results = model.get_current_visuals(entire_batch=True)
train_psnrs = [
util.calculate_psnr(util.tensor2img(cur_train_results['SR'][im_num], out_type=np.float32) * 255,
util.tensor2img(cur_train_results['HR'][im_num], out_type=np.float32) * 255) for
im_num in range(len(cur_train_results['SR']))]
#Save latest training batch output:
save_img_path = os.path.join(os.path.join(opt['path']['val_images']),
'{:d}_Tr_PSNR{:.3f}.png'.format(model.gradient_step_num, np.mean(train_psnrs)))
util.save_img(np.clip(np.concatenate((np.concatenate([util.tensor2img(cur_train_results['HR'][im_num], out_type=np.float32) * 255 for im_num in
range(len(cur_train_results['SR']))],0), np.concatenate([util.tensor2img(cur_train_results['SR'][im_num], out_type=np.float32) * 255 for im_num in
range(len(cur_train_results['SR']))],0)), 1), 0, 255).astype(np.uint8), save_img_path)
Z_latent = [0]+([-1,1] if (opt['network_G']['latent_input'] and model.num_latent_channels>0) else [])
print_rlt['psnr'],print_rlt['niqe'] = 0,0
model.im_collages = []
for cur_Z in Z_latent:
sr_images = model.perform_validation(data_loader=val_loader,cur_Z=cur_Z,print_rlt=print_rlt,first_eval=save_GT_HR,save_images=True)
if logger.use_tb_logger:
logger.tb_logger.log_images('validation_Z%.2f'%(cur_Z), [im[:,:,[2,1,0]] for im in sr_images], model.gradient_step_num)
if save_GT_HR: # Save GT Uncomp images
save_GT_HR = False
util.prune_old_files(cur_step=model.gradient_step_num, folder=model.opt['path']['val_images'],
saving_freq=opt['train']['val_save_freq'],name_pattern='^(\d)+'+('_Z' if len(Z_latent)>1 else '')+'.*PSNR.*.png$')
print_rlt['psnr'] /= len(Z_latent)
# print_rlt['niqe'] /= len(Z_latent)
model.log_dict['psnr_val'].append((model.gradient_step_num,print_rlt['psnr']))
# model.log_dict['niqe_val'].append((model.gradient_step_num,print_rlt['niqe']))
if len(Z_latent)>1:
print_rlt['per_pix_STD'] = np.mean(np.std(np.stack(model.im_collages, 0), 0))
model.log_dict['per_pix_STD_val'].append((model.gradient_step_num,print_rlt['per_pix_STD']))
else:
print('Skipping validation because generator is unchanged')
time_elapsed = time.time() - start_time
# Save to log
print_rlt['model'] = opt['model']
print_rlt['epoch'] = epoch
print_rlt['iters'] = model.gradient_step_num
print_rlt['time'] = time_elapsed
model.display_log_figure()
logger.print_format_results('val', print_rlt,keys_ignore_list=IGNORED_KEYS_LIST)
print('-----------------------------------')
# update learning rate
if not_within_batch:
lr_too_low = model.update_learning_rate(model.gradient_step_num)
if lr_too_low:
print('Stopping training because LR is too low')
break
print('Saving the final model.')
model.save(model.gradient_step_num)
print('End of training.')
if __name__ == '__main__':
# # OpenCV get stuck in transform when used in DataLoader
# # https://github.com/pytorch/pytorch/issues/1838
# # However, cause problem reading lmdb
# import torch.multiprocessing as mp
# mp.set_start_method('spawn', force=True)
main()