-
Notifications
You must be signed in to change notification settings - Fork 0
/
train_model.py
161 lines (127 loc) · 5.98 KB
/
train_model.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
import argparse
import os
import numpy as np
import math
import torch
import torch.nn as nn
import torch.nn.functional as F
from network import EnhancerModel
from losses import *
import datasets
import PIL
from tqdm import tqdm
from torch.utils.tensorboard import SummaryWriter
from datetime import datetime
import torchvision.transforms as transforms
from torchvision.utils import save_image
from datetime import datetime
def train(opt):
torch.manual_seed(0)
device = 'cuda' if torch.cuda.is_available() else 'cpu'
logger = SummaryWriter(os.path.join(opt.output_dir, 'logs'))
if device == 'cpu':
print("cpu is the device")
return
# device='cpu'
model = EnhancerModel.DceNet().to(device)
optimizer = torch.optim.Adam(model.parameters(), weight_decay = 0.0001,
lr = 1e-4)
dataloader = datasets.get_dataset(
dataset_path = opt.dataset_path,
img_size = opt.img_size,
batch_size=opt.batch_size
)
total_progress_bar = tqdm(
total = opt.n_epochs,
desc = 'Total progress',
dynamic_ncols = True
)
steps = 0
for ep in range(1, opt.n_epochs+1):
total_progress_bar.update(1)
losses = []
losses_spa = []
losses_exp = []
losses_col = []
losses_ill = []
# start_s = datetime.now()
for imgs in tqdm(dataloader):
# for i, imgs in enumerate(dataloader):
end_s = datetime.now()
# print("time take is {}".format((end_s - start_s).total_seconds()))
start = datetime.now()
imgs = imgs.to(device).float()
maps, enhanced_image = model(imgs)
end = datetime.now()
# print("time take for forward prediction {}".format((end - start).total_seconds()))
start = datetime.now()
loss_spt = spatial_consistency_loss(imgs, enhanced_image, local_region=opt.spt_local_region)
end = datetime.now()
# print("time take for spatial_consistency_loss {}".format((end - start).total_seconds()))
start = datetime.now()
loss_exp = exposure_control_loss(enhanced_image, E = opt.E, local_region = opt.exp_local_region)
end = datetime.now()
# print("time take for exposure_control_loss {}".format((end - start).total_seconds()))
start = datetime.now()
loss_col = color_consistency_loss(enhanced_image)
end = datetime.now()
# print("time take for color_consistency_loss {}".format((end - start).total_seconds()))
start = datetime.now()
loss_ill = illumination_smoothness_loss(maps)
end = datetime.now()
# print("time take for illumination_smoothness_loss {}".format((end - start).total_seconds()))
start = datetime.now()
total_loss = opt.spa_weight*loss_spt + opt.exp_weight*loss_exp + opt.color_weight*loss_col + opt.ill_weight*loss_ill
logger.add_scalar('spatial consistency loss', loss_spt.item(), steps)
logger.add_scalar('exposure control loss', loss_exp.item(), steps)
logger.add_scalar('color consistency loss', loss_col.item(), steps)
logger.add_scalar('illumination smoothness loss', loss_ill.item(), steps)
logger.add_scalar('total loss', total_loss.item(), steps)
losses.append(total_loss.item())
losses_spa.append(loss_spt.item())
losses_exp.append(loss_exp.item())
losses_col.append(loss_col.item())
losses_ill.append(loss_ill.item())
end = datetime.now()
# print("time take for logging {}".format((end - start).total_seconds()))
start = datetime.now()
optimizer.zero_grad()
total_loss.backward()
torch.nn.utils.clip_grad_norm_(model.parameters(), 0.1)
optimizer.step()
end = datetime.now()
# print("time take for grad backward {}".format((end - start).total_seconds()))
steps += 1
start_s = datetime.now()
tqdm.write(f"[Epoch : {ep+1}/{opt.n_epochs}][total_loss : {np.mean(losses)}] [losses_spa : {np.mean(losses_spa)}] [losses_exp : {np.mean(losses_exp)}] [losses_col : {np.mean(losses_col)}] [losses_ill : {np.mean(losses_ill)}]")
# print("saving test image")
# torch.cuda.empty_cache()
# test_image = PIL.Image.open('D:\\spring2023\\748\\project\\data\\Dataset_Part1\\Dataset_Part1\\318\\1.JPG')
# test_image = transforms.ToTensor()(test_image).to(device)
# with torch.no_grad():
# _, en_img = model(test_image)
# save_image(en_img, 'output_{}.jpg'.format(ep))
if ep % 5 == 0:
torch.save(model.state_dict(), opt.output_dir + '/enhancer_model_v_{}.pth'.format(ep))
torch.cuda.empty_cache()
torch.save(model.state_dict(), opt.output_dir + '/enhancer_model_final.pth')
if __name__ == '__main__':
parser = argparse.ArgumentParser()
parser.add_argument("--n_epochs", type=int, default=30)
parser.add_argument("--dataset_path", type=str, default = 'data/')
parser.add_argument("--img_size", type=int, default=256)
parser.add_argument("--batch_size", type=int, default=8)
parser.add_argument("--spt_local_region", type=int, default=4)
parser.add_argument("--exp_local_region", type=int, default=16)
parser.add_argument("--exp_weight", type=int, default=10)
parser.add_argument("--color_weight", type=int, default=5)
parser.add_argument("--ill_weight", type=int, default=200)
parser.add_argument("--spa_weight", type=int, default=1)
parser.add_argument("--E", type=int, default=0.6)
parser.add_argument("--output_dir", type=str, default = 'models/')
opt = parser.parse_args()
print(opt)
start = datetime.now()
train(opt)
end = datetime.now()
print("total time take is {}".format((end - start).total_seconds()/60.0))