-
Notifications
You must be signed in to change notification settings - Fork 9
/
main.py
236 lines (196 loc) · 8.43 KB
/
main.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
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
import torch.utils.data as data
import torch.optim as optim
from torch.autograd import Variable
import torch.optim.lr_scheduler as lr_scheduler
import torch.nn as nn
import glob
from tqdm import tqdm
import os
from PIL import Image
from utils.dataset import GIANA
import utils.transforms as t
import utils.co_transforms as cot
from models.unet import (UNet, UNetRes)
from models.fcn import (FCN32, FCN16, FCN8)
from models.pspnet import PSPNet
from models.segnet import SegNet
from models.gcn import ResnetGCN
from models.fusionnet import FusionNet
from models.linknet import LinkNet
from models.duc import (ResentDUC, ResentDUCHDC)
from models.discriminator import Discriminator
from utils.avg import AverageMeter
from utils.evaluation import Evaluation
from utils.visualize import Dashboad
from utils.losses import *
from utils.experiments import Experiments
from utils.tools import morph_postprocess
from settings import get_arguments
import sys
import itertools
def load_data(args):
normalize = t.Normalize(mean=[0.445, 0.287, 0.190], std=[0.31, 0.225, 0.168])
im_transform = t.Compose([t.ToTensor(), normalize])
# Use the following code fo co_transformations e.g. random rotation or random flip etc.
# co_transformer = cot.Compose([cot.RandomRotate(45)])
dsetTrain = GIANA(args.imgdir, args.gtdir, input_size=(args.input_width, args.input_height) ,train=True, transform=im_transform, co_transform=None, target_transform=t.ToLabel())
train_data_loader = data.DataLoader(dsetTrain, batch_size=args.batch_size, shuffle=True, num_workers=args.num_workers)
dsetVal = GIANA(args.imgdir, args.gtdir, train=False, transform=im_transform, co_transform=None,
target_transform=t.ToLabel())
val_data_loader = data.DataLoader(dsetVal, batch_size=args.batch_size, shuffle=False,
num_workers=args.num_workers)
return train_data_loader, val_data_loader
def train(args, model):
board = Dashboad(args.visdom_port)
tr_losses = AverageMeter()
tLoader, vLoader = load_data(args)
criterion = nn.BCELoss()
optimizer = optim.SGD(model.parameters(), lr=args.lr, weight_decay=args.weight_decay, momentum=0.99)
scheduler = lr_scheduler.StepLR(optimizer, step_size=30, gamma=0.1)
for epoch in range(1, args.num_epochs+1):
scheduler.step()
if epoch == 1:
tr_loss, _ , _ = evaluate(args, model, criterion, tLoader)
vl_loss, vl_jacc, vl_dice = evaluate(args, model, criterion, vLoader)
# Draw the loss curves
win = None
win = board.loss_curves([tr_loss, vl_loss], epoch, win=win)
print('[Initial TrainLoss: {0:.4f}]'
'\t[Initial ValidationLoss: {1:.4f}]'
'\t[Initial ValidationJaccard: {2:.4f}]'
'\t[Initial ValidationDice: {3:.4f}]'.format(tr_loss, vl_loss, vl_jacc, vl_dice))
print('----------------------------------------------------------------------------------------------------'
'--------------')
for step, (images, labels) in enumerate(tLoader):
model.train(True)
if args.cuda:
images = images.cuda()
labels = labels.cuda()
criterion = criterion.cuda()
inputs = Variable(images)
targets = Variable(labels)
optimizer.zero_grad()
outputs = model(inputs)
loss = criterion(outputs, targets)
loss.backward()
optimizer.step()
tr_losses.update(loss.data.cpu().numpy())
if epoch % args.log_step == 0:
vl_loss, vl_jacc, vl_dice = evaluate(args, model, criterion, vLoader)
print('[Epoch: {0:02}/{1:02}]'
'\t[TrainLoss: {2:.4f}]'
'\t[ValidationLoss: {3:.4f}]'
'\t[ValidationJaccard: {4:.4f}]'
'\t[ValidationDice: {5:.4f}]'.format(epoch, args.num_epochs, tr_losses.avg, vl_loss, vl_jacc,
vl_dice)),
filename = "weights/{0}-{1:02}.pth".format(args.model, epoch)
torch.save(model.state_dict(), filename)
print(' [Snapshot]')
else:
vl_loss, vl_jacc, vl_dice = evaluate(args, model, criterion, vLoader)
print('[Epoch: {0:02}/{1:02}]'
'\t[TrainLoss: {2:.4f}]'
'\t[ValidationLoss: {3:.4f}]'
'\t[ValidationJaccard: {4:.4f}]'
'\t[ValidationDice: {5:.4f}]'.format(epoch, args.num_epochs, tr_losses.avg, vl_loss, vl_jacc,
vl_dice))
# --- Update the loss curves ---
win = board.loss_curves([tr_losses.avg, vl_loss], epoch, win=win)
def evaluate(args, model, criterion, val_loader):
model.eval()
losses = AverageMeter()
jaccars = AverageMeter()
dices = AverageMeter()
eva = Evaluation()
for i, (images, labels) in enumerate(val_loader):
if args.cuda:
images = images.cuda()
labels = labels.cuda()
criterion = criterion.cuda()
images = Variable(images)
labels = Variable(labels)
outputs = model(images)
loss = criterion(outputs, labels)
losses.update(loss.data.cpu().numpy())
jacc_index = eva.jaccard_similarity_coefficient(outputs.cpu().data.numpy().squeeze(),
labels.cpu().data.numpy())
dice_index = eva.dice_coefficient(outputs.cpu().data.numpy().squeeze(),
labels.cpu().data.numpy())
jaccars.update(jacc_index)
dices.update(dice_index)
return losses.avg, jaccars.avg, dices.avg
def eval(args, model):
if not os.path.exists(args.savedir):
os.makedirs(args.savedir)
listImgFiles = [k.split('/')[-1].split('.')[0] for k in glob.glob(os.path.join(args.imgdir, '*.bmp'))]
for currFile in tqdm(listImgFiles):
img = Image.open(os.path.join(args.imgdir, currFile + '.bmp'))
img = t.ToTensor()(img)
if args.cuda:
img.cuda()
output = model(Variable(img, volatile=True).unsqueeze(0))
output = t.ToPILImage()(output[0].cpu().data)
newfilename = os.path.join(args.savedir, currFile + '_' + args.model + '.bmp')
output.save(newfilename, 'BMP')
def make_model(args):
Net = None
if args.model == 'unet':
Net = UNet
elif args.model == 'unet_res':
Net = UNetRes
elif args.model == 'fcn32':
Net = FCN32
elif args.model == 'fcn16':
Net = FCN16
elif args.model == 'fcn8':
Net = FCN8
elif args.model == 'pspnet':
Net = PSPNet
elif args.model == 'segnet':
Net = SegNet
elif args.model == 'fusionnet':
Net = FusionNet
elif args.model == 'duc':
Net = ResentDUC
elif args.model == 'duchdc':
Net = ResentDUCHDC
elif args.model == 'gcn':
Net = ResnetGCN
elif args.model == 'linknet':
Net = LinkNet
assert Net is not None, 'model {args.model} is not available'
model = Net(1)
if args.cuda:
nGPUs = torch.cuda.device_count()
model = torch.nn.DataParallel(model, device_ids=range(nGPUs)).cuda()
model = model.cuda()
if args.state:
model.load_state_dict(torch.load('weights/' + args.state))
# In case to use weights trained on multiple GPUs in CPU mode
else:
checkpoint = torch.load('./weights/' + args.state)
state_dict = checkpoint['state_dict']
from collections import OrderedDict
new_state_dict = OrderedDict()
for k, v in state_dict.items():
name = k[0:9] + k[16:] # remove `module.`
if k[0] == 'f':
new_state_dict[name] = v
else:
new_state_dict[k] = v
model.load_state_dict(new_state_dict)
model = model.cpu()
return model
def main(args):
torch.backends.cudnn.enabled = False
model = make_model(args)
if args.mode == 'train':
train(args, model)
elif args.mode == 'eval':
eval(args, model)
if __name__ == '__main__':
args = get_arguments(sys.argv[1:])
if args.mode == 'crossval':
params = list(itertools.product(args.batch_size, args.optimizer, args.momentum, args.lr))
args.batch_size, args.optimizer, args.momentum, args.lr = params[args.settings_id]
main(args)