-
Notifications
You must be signed in to change notification settings - Fork 0
/
run.py
148 lines (131 loc) · 5.49 KB
/
run.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
#!/usr/bin/env python
# encoding: utf-8
import argparse
import torch
import torch.nn as nn
from torchvision import transforms
import numpy as np
import os, sys
from args import *
#from net import *
from net_origin import *
from data_loader import *
from utils import *
from PIL import Image
from test import *
def run_net(args):
os.environ['CUDA_VISIBLE_DEVICES'] = args.gpu_list
# Create model directory
if not os.path.exists(args.save_folder):
os.makedirs(args.save_folder)
transform = transforms.Compose([
transforms.ToTensor(),
#transforms.Normalize((0.4850, 0.4580, 0.4077), (1.0, 1.0, 1.0))
#transforms.Normalize((123.68, 116.78, 103.94), (1.0, 1.0, 1.0))
])
block_cnt = [int(c) for c in args.blocks.split(",") ]
net = FancyNet(block_cnt)
#print "##### net:",net
# net = ResNet(Bottleneck, [3, 4, 6, 3])
if torch.cuda.is_available():
print 'USE CUDA : net'
net = net.cuda()
#print net.keys()
print 'before resume'
if args.resume_flag == 1 and args.resume_path is not None:
print 'resume in', args.resume_path
trained_model = torch.load(args.resume_path)
'''
(key, value) = trained_model.popitem()
print key, 'is remove from trained_model'
(key, value) = trained_model.popitem()
print key, 'is remove from trained_model'
'''
# print trained_model.keys()
net.load_state_dict(trained_model)
print 'resume_successs'
else:
print 'no resume path, start from random initialization'
if args.phase == "test":
print 'TEST phase'
net = net.eval()
test(args, net, transform)
else:
print 'TRAIN phase'
train(args, net, transform)
def train(args, net, transform):
loss_ratio = args.loss_ratio
#shuffle = True
shuffle = False
data_loader = get_loader(args.root_folder, args.bg_folder,
transform = transform, batch_size = args.batch_size,
shuffle = shuffle, num_workers = args.num_workers)
optimizer = torch.optim.Adam(net.parameters(), lr = args.learning_rate, weight_decay = 0.0005)
step_change_lr_1 = max(1, (args.start_epoch*2 + args.end_epoch*3)//5 - 1)
step_change_lr_2 = max(1, (args.start_epoch*1 + args.end_epoch*4)//5 - 1)
step_save_model = max(1, (args.end_epoch - args.start_epoch)//10)
step_change_loss_ratio = (args.end_epoch - args.start_epoch)//5
for epoch in range(args.start_epoch, args.end_epoch):
print "EPOCH", epoch
for i, (images, score_maps, geo_maps, train_masks) in enumerate(data_loader):
#print epoch, i
images = to_cuda(images)
y_scores = to_cuda(score_maps)
y_geos = to_cuda(geo_maps)
y_masks = to_cuda(train_masks)
time0 = time.time()
p_scores, p_geos, p_angle = net(images)
# time1 = time.time()
#print "BE LOSS", y_masks.size(), torch.sum(y_masks[0])
#print "be LOSS", y_scores.size(), torch.sum(y_scores[0])
seg_loss, dice_loss, balance_loss = EastLoss(p_scores, p_geos, p_angle, y_scores, y_geos, y_masks)
# time2 = time.time()
# print('net_time :', time1 - time0)
# print('loss_time: ', time2 - time1)
# print('total_time: ', time2 - time0)
# loss = seg_loss + loss_ratio * dice_loss / 2.0 + loss_ratio * balance_loss
loss = seg_loss + loss_ratio * dice_loss
# loss = seg_loss + loss_ratio * dice_loss / 2.0 + loss_ratio * balance_loss
# loss = balance_loss
optimizer.zero_grad()
loss.backward()
optimizer.step()
if (i + 1) % 10 == 0:
print (get_time_by_sec(False) + 'Epoch [%d/%d], Iter [%d/%d] , Seg: %.6f, Dice: %.6f, Blan: %.6f, Total: %.6f'%(epoch, args.end_epoch, i + 1, data_loader.__len__(), seg_loss.data[0], dice_loss.data[0], balance_loss.data[0], loss.data[0]))
if epoch == step_change_lr_1:
print 'change lr to', args.learning_rate * 0.5
optimizer = torch.optim.Adam(net.parameters(), lr = args.learning_rate * 0.5, weight_decay = 0.0005)
elif epoch == step_change_lr_2:
print 'change lr to', args.learning_rate * 0.25
optimizer = torch.optim.Adam(net.parameters(), lr = args.learning_rate * 0.25, weight_decay = 0.0005)
#if i == 1:
# break
#break
# if (epoch + 1 - args.start_epoch) % step_change_loss_ratio == 0 and loss_ratio > 0.001:
# loss_ratio = loss_ratio / 2
# if (epoch + 1 - args.start_epoch) % step_change_loss_ratio == 0 and loss_ratio > 0.1:
# loss_ratio = loss_ratio - 0.2
if (epoch + 1 - args.start_epoch) % step_save_model == 0:
save_path = args.save_folder
if save_path[-1] == '/':
save_path = save_path[:-1] + "_"
else:
save_path += "_"
save_path += str(epoch + 1) + ".ckpt"
save_path = os.path.join(args.save_folder, save_path)
print 'save_path: ', save_path
torch.save(net.state_dict(), save_path)
if __name__ == "__main__":
args = load_args()
print args
run_net(args)
'''
if args.phase == "train":
print 'TRAIN phase'
train(args)
elif args.phase == "test":
print 'TEST phase'
test(args)
else:
print "not known phase"
'''