forked from HaoranZhao1988/DDAD
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathtrain_teacher_seg.py
152 lines (131 loc) · 6.96 KB
/
train_teacher_seg.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
from __future__ import print_function
import argparse
import torch
import torch.nn.functional as F
import torch.optim as optim
from torchvision import datasets, transforms
import torchvision
import network
from dataloader import get_dataloader
from utils.stream_metrics import StreamSegMetrics
from utils.visualizer import VisdomPlotter
from utils.misc import pack_images, denormalize
from collections import OrderedDict
from utils import focal_loss
import numpy as np
import random
vp = None
def train(args, model, device, train_loader, optimizer, epoch):
model.train()
for batch_idx, (data, target) in enumerate(train_loader):
data, target = data.to(device), target.to(device, dtype=torch.long)
optimizer.zero_grad()
output = model(data)
loss = focal_loss(output, target, gamma=2, ignore_index=255) #focal_loss(output, target, gamma=2)#F.cross_entropy(output, target, ignore_index=255)
loss.backward()
optimizer.step()
if batch_idx % args.log_interval == 0:
print('Train Epoch: {} [{}/{} ({:.0f}%)]\tLoss: {:.6f}'.format(
epoch, batch_idx * len(data), len(train_loader.dataset),
100. * batch_idx / len(train_loader), loss.item()))
def test(args, model, device, test_loader):
model.eval()
test_loss = 0
correct = 0
seg_metrics = StreamSegMetrics(args.num_classes)
with torch.no_grad():
for i, (data, target) in enumerate(test_loader):
data, target = data.to(device), target.to(device, dtype=torch.long)
output = model(data)
seg_metrics.update(output.max(1)[1].detach().cpu().numpy().astype('uint8'), target.detach().cpu().numpy().astype('uint8'))
# if i==0:
# vp.add_image( 'input', pack_images( ((data+1)/2).clamp(0, 1.0).cpu().numpy() ) )
# vp.add_image( 'target', pack_images( test_loader.dataset.decode_target(target.cpu().numpy()), channel_last=True ).astype('uint8') )
# vp.add_image( 'pred', pack_images( test_loader.dataset.decode_target(output.max(1)[1].detach().cpu().numpy().astype('uint8')), channel_last=True ).astype('uint8') )
results = seg_metrics.get_results()
print('\nTest set: Acc= %.6f, mIoU: %.6f\n'%(results['Overall Acc'],results['Mean IoU']))
return results
def get_model(args):
if args.model.lower()=='deeplabv3_resnet50':
return network.segmentation.deeplabv3.deeplabv3_resnet50(num_classes=args.num_classes, dropout_p=0.5, pretrained_backbone=True)
elif args.model.lower()=='segnet_vgg19':
return network.segmentation.segnet.SegNetVgg19(args.num_classes, pretrained_backbone=True)
elif args.model.lower()=='segnet_vgg16':
return network.segmentation.segnet.SegNetVgg16(args.num_classes, pretrained_backbone=True)
elif args.model.lower()=='segnet_vgg13':
return network.segmentation.segnet.SegNetVgg13(args.num_classes, pretrained_backbone=True)
def main():
# Training settings
parser = argparse.ArgumentParser()
parser.add_argument('--num_classes', type=int, default=21)
parser.add_argument('--batch_size', type=int, default=16, metavar='N',
help='input batch size for training (default: 64)')
parser.add_argument('--test_batch_size', type=int, default=16, metavar='N',
help='input batch size for testing (default: 1000)')
parser.add_argument('--epochs', type=int, default=300, metavar='N',
help='number of epochs to train (default: 10)')
parser.add_argument('--lr', type=float, default=0.01, metavar='LR',
help='learning rate (default: 0.1)')
parser.add_argument('--data_root', type=str, default='/root/data/unpacked/')
parser.add_argument('--dataset', type=str, default='voc2012', choices=['camvid', 'nyuv2', 'cityscapes','voc2012'],
help='dataset name (default: camvid)')
parser.add_argument('--model', type=str, default='deeplabv3_resnet50', choices=['deeplabv3_resnet50', 'segnet_vgg19', 'segnet_vgg16'],
help='model name (default: deeplabv3_resnet50)')
parser.add_argument('--weight_decay', type=float, default=1e-4)
parser.add_argument('--gamma', type=float, default=0.1)
parser.add_argument('--momentum', type=float, default=0.9, metavar='M',
help='SGD momentum (default: 0.9)')
parser.add_argument('--no_cuda', action='store_true', default=False,
help='disables CUDA training')
parser.add_argument('--seed', type=int, default=1, metavar='S',
help='random seed (default: 1)')
parser.add_argument('--step_size', type=int, default=100, metavar='S',
help='random seed (default: 1)')
parser.add_argument('--ckpt', type=str, default=None)
parser.add_argument('--log_interval', type=int, default=10, metavar='N',
help='how many batches to wait before logging training status')
parser.add_argument('--test_only', action='store_true', default=False)
parser.add_argument('--download', action='store_true', default=False)
parser.add_argument('--scheduler', action='store_true', default=False)
args = parser.parse_args()
use_cuda = not args.no_cuda and torch.cuda.is_available()
# torch.manual_seed(args.seed)
# torch.cuda.manual_seed(args.seed)
# np.random.seed(args.seed)
# random.seed(args.seed)
# torch.backends.cudnn.deterministic = True
# torch.backends.cudnn.benchmark = False
device = torch.device("cuda" if use_cuda else "cpu")
kwargs = {'num_workers': 1, 'pin_memory': True} if use_cuda else {}
print(args)
# global vp
# vp = VisdomPlotter('15550', 'teacher-seg-%s'%args.dataset)
train_loader, test_loader = get_dataloader(args)
model = get_model(args)
if args.ckpt is not None:
model.load_state_dict( torch.load( args.ckpt ) )
model = model.to(device)
if torch.cuda.device_count() > 1:
model = torch.nn.DataParallel(model)
optimizer = optim.SGD(model.parameters(), lr=args.lr, weight_decay=args.weight_decay, momentum=args.momentum)
best_result = 0
if args.scheduler:
scheduler = optim.lr_scheduler.StepLR(optimizer, args.step_size, gamma=args.gamma)
if args.test_only:
results = test(args, model, device, test_loader)
return
for epoch in range(1, args.epochs + 1):
if args.scheduler:
scheduler.step()
print("Lr = %.6f"%(optimizer.param_groups[0]['lr']))
print("train")
train(args, model, device, train_loader, optimizer, epoch)
print("val")
results = test(args, model, device, test_loader)
# vp.add_scalar('mIoU', epoch, results['Mean IoU'])
if results['Mean IoU']>best_result:
best_result = results['Mean IoU']
torch.save(model.state_dict(),"checkpoint/teacher/%s-%s-3.pt"%(args.dataset, args.model))
print("Best mIoU=%.6f"%best_result)
if __name__ == '__main__':
main()