-
Notifications
You must be signed in to change notification settings - Fork 0
/
train_B.py
177 lines (134 loc) · 7.65 KB
/
train_B.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
import torch
from torch.autograd import Variable
import torch.nn as nn
import torch.nn.functional as F
from torch.utils.data import Dataset, DataLoader
from torchvision import transforms, utils
import torch.optim as optim
import os, argparse
from ColorJitter import ColorJitter
from datetime import datetime
from dataset import RescaleT
from dataset import RandomCrop
from dataset import ToTensorLab
from dataset import SalObjDataset
from model.PAFR import Net
from src.G_smooth import SSIM
from src.G_crf import GatedCRF
os.environ["CUDA_VISIBLE_DEVICES"] = '0'
parser = argparse.ArgumentParser()
parser.add_argument('--epoch', type=int, default=60, help='epoch number')
parser.add_argument('--lr', type=float, default=1e-4, help='learning rate')
parser.add_argument('--batchsize', type=int, default=1, help='training batch size')
parser.add_argument('--clip', type=float, default=0.5, help='gradient clipping margin')
parser.add_argument('--decay_rate', type=float, default=0.9, help='decay rate of learning rate')
parser.add_argument('--decay_epoch', type=int, default=40, help='every n epochs decay learning rate')
opt = parser.parse_args()
model = Net(3)
model.cuda()
optimizer = optim.AdamW(model.parameters(), lr=opt.lr, betas=(0.9, 0.999), eps=1e-08, weight_decay=0)
loss_lsc_kernels_desc_defaults = [{"weight": 1, "xy": 6, "rgb": 0.1}]
loss_lsc_radius = 5
l = 0.3
CJ = ColorJitter(brightness=0.2, contrast=0.2, saturation=0.2, hue=0.1)
bce_loss = nn.BCELoss(size_average=True)
ssim_loss = SSIM(window_size=11, size_average=True)
loss_lsc = GatedCRF().cuda()
tra_image_dir = './train_data/EORRSD-S/img/'
tra_gt_dir = './train_data/EORRSD-S/gt/'
tra_mask_dir = './train_data/EORRSD-S/mask/'
tra_img_name_list = [tra_image_dir + f for f in os.listdir(tra_image_dir) if f.endswith('.jpg')]
tra_gt_name_list = [tra_gt_dir + f for f in os.listdir(tra_gt_dir) if f.endswith('.png')]
tra_mask_name_list = [tra_mask_dir + f for f in os.listdir(tra_mask_dir) if f.endswith('.png')]
print("---")
print("train images: ", len(tra_img_name_list))
print("---")
salobj_dataset = SalObjDataset(
img_name_list=tra_img_name_list,
gt_name_list=tra_gt_name_list,
mask_name_list=tra_mask_name_list,
transform=transforms.Compose([
RescaleT(256),
RandomCrop(224),
ToTensorLab(flag=0),
ColorJitter(brightness=0.2, contrast=0.2, saturation=0.2, hue=0.1)]))
salobj_dataloader = DataLoader(salobj_dataset, batch_size=opt.batchsize, shuffle=True, num_workers=0)
train_num = len(salobj_dataloader)
def clip_gradient(optimizer, grad_clip):
for group in optimizer.param_groups:
for param in group['params']:
if param.grad is not None:
param.grad.data.clamp_(-grad_clip, grad_clip)
def adjust_lr(optimizer, init_lr, epoch, decay_rate=0.1, decay_epoch=30):
decay = decay_rate ** (epoch // decay_epoch)
for param_group in optimizer.param_groups:
param_group['lr'] *= decay
def sscloss(x, y, alpha):
ssim_out = 1 - ssim_loss(x, y)
l1_loss = torch.mean(torch.abs(x-y))
loss_ssc = alpha*ssim_out + (1-alpha)*l1_loss
return loss_ssc
def train(salobj_dataloader, model, optimizer, epoch):
model.train()
for i, data in enumerate(salobj_dataloader, start=1):
optimizer.zero_grad()
images, gts, masks = data['image'], data['gt'], data['mask']
images_colorJ = CJ(images)
images = images.type(torch.FloatTensor)
gts = gts.type(torch.FloatTensor)
masks = masks.type(torch.FloatTensor)
images_colorJ = Variable(images_colorJ.cuda(), requires_grad=False)
images = Variable(images.cuda(), requires_grad=False)
gts = Variable(gts.cuda(), requires_grad=False)
masks = Variable(masks.cuda(), requires_grad=False)
img_size = images.size(2) * images.size(3) * images.size(0)
ratio = img_size / torch.sum(masks)
#############################################################################################
image_scaleCJ = F.interpolate(images_colorJ, scale_factor=0.3, mode='bilinear', align_corners=True) # [1, 3, 67, 67]
d1, d2, d3, d4, d5, d6, d7 = model(images)
d1_s, d2_s, d3_s, d4_s, d5_s, d6_s, d7_s = model(image_scaleCJ)
d1_scale = F.interpolate(d1, size=d1_s.size()[2:], mode='bilinear', align_corners=True) # [1, 1, 96, 96]
d2_scale = F.interpolate(d2, size=d2_s.size()[2:], mode='bilinear', align_corners=True)
loss_ssc1 = sscloss(d1_s, d1_scale, 0.85)
loss_ssc2 = sscloss(d2_s, d2_scale, 0.85)
loss_ssc = loss_ssc1 + loss_ssc2
image_ = F.interpolate(images, scale_factor=0.25, mode='bilinear', align_corners=True)
sample = {'rgb': image_}
d1_ = F.interpolate(d1, scale_factor=0.25, mode='bilinear', align_corners=True)
loss1_lsc = loss_lsc(d1_, loss_lsc_kernels_desc_defaults, loss_lsc_radius, sample, image_.shape[2], image_.shape[3])['loss']
loss1 = ratio*bce_loss(d1*masks, gts*masks) + l * loss1_lsc + loss_ssc
d2_ = F.interpolate(d2, scale_factor=0.25, mode='bilinear', align_corners=True)
loss2_lsc = loss_lsc(d2_, loss_lsc_kernels_desc_defaults, loss_lsc_radius, sample, image_.shape[2], image_.shape[3])['loss']
loss2 = ratio*bce_loss(d2*masks, gts*masks) + l * loss2_lsc
d3_ = F.interpolate(d3, scale_factor=0.25, mode='bilinear', align_corners=True)
loss3_lsc = loss_lsc(d3_, loss_lsc_kernels_desc_defaults, loss_lsc_radius, sample, image_.shape[2], image_.shape[3])['loss']
loss3 = ratio*bce_loss(d3*masks, gts*masks) + l * loss3_lsc
d4_ = F.interpolate(d4, scale_factor=0.25, mode='bilinear', align_corners=True)
loss4_lsc = loss_lsc(d4_, loss_lsc_kernels_desc_defaults, loss_lsc_radius, sample, image_.shape[2], image_.shape[3])['loss']
loss4 = ratio*bce_loss(d4*masks, gts*masks) + l * loss4_lsc
d5_ = F.interpolate(d5, scale_factor=0.25, mode='bilinear', align_corners=True)
loss5_lsc = loss_lsc(d5_, loss_lsc_kernels_desc_defaults, loss_lsc_radius, sample, image_.shape[2], image_.shape[3])['loss']
loss5 = ratio*bce_loss(d5*masks, gts*masks) + l * loss5_lsc
d6_ = F.interpolate(d6, scale_factor=0.25, mode='bilinear', align_corners=True)
loss6_lsc = loss_lsc(d6_, loss_lsc_kernels_desc_defaults, loss_lsc_radius, sample, image_.shape[2], image_.shape[3])['loss']
loss6 = ratio*bce_loss(d6*masks, gts*masks) + l * loss6_lsc
d7_ = F.interpolate(d7, scale_factor=0.25, mode='bilinear', align_corners=True)
loss7_lsc = loss_lsc(d7_, loss_lsc_kernels_desc_defaults, loss_lsc_radius, sample, image_.shape[2], image_.shape[3])['loss']
loss7 = ratio*bce_loss(d7*masks, gts*masks) + l * loss7_lsc
loss = loss1 + loss2 + loss3 + loss4 + loss5 + loss6 + loss7
loss.backward()
clip_gradient(optimizer, opt.clip)
optimizer.step()
if i % 10 == 0 or i == train_num:
print('{} Epoch [{:03d}/{:03d}], Step [{:04d}/{:04d}], refine_loss: {:0.4f}, total_loss: {:0.4f}'.
format(datetime.now(), epoch, opt.epoch, i, train_num, loss1.data, loss.data))
model_dir = "./saved_models/"
if not os.path.exists(model_dir):
os.makedirs(model_dir)
if epoch % 10 == 0:
torch.save(model.state_dict(), model_dir + 'mode_B' + '_%d' % epoch + '.pth')
print('Start Training!')
if __name__ == '__main__':
for epoch in range(1, opt.epoch+1):
adjust_lr(optimizer, opt.lr, epoch, opt.decay_rate, opt.decay_epoch)
train(salobj_dataloader, model, optimizer, epoch)