-
Notifications
You must be signed in to change notification settings - Fork 5
/
train.py
207 lines (145 loc) · 7.61 KB
/
train.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
import os, argparse, time
import numpy as np
import torch
import torch.nn as nn
from torchvision import transforms
import torch.backends.cudnn as cudnn
import dataset
import multi_channel_resnet34_hyper
from scipy import stats
from scipy.optimize import curve_fit
def logistic_func(X, bayta1, bayta2, bayta3, bayta4):
logisticPart = 1 + np.exp(np.negative(np.divide(X - bayta3, np.abs(bayta4))))
yhat = bayta2 + np.divide(bayta1 - bayta2, logisticPart)
return yhat
def fit_function(y_label, y_output):
beta = [np.max(y_label), np.min(y_label), np.mean(y_output), 0.5]
popt, _ = curve_fit(logistic_func, y_output, \
y_label, p0=beta, maxfev=100000000)
y_output_logistic = logistic_func(y_output, *popt)
return y_output_logistic
def parse_args():
"""Parse input arguments. """
parser = argparse.ArgumentParser(description="No reference 360 degree image quality assessment.")
parser.add_argument('--gpu', dest='gpu_id', help="GPU device id to use [0]", default=0, type=int)
parser.add_argument('--num_epochs', dest='num_epochs', help='Maximum number of training epochs.',
default=5, type=int)
parser.add_argument('--lr', dest='lr', help='learning rate.',
default=0.0001, type=float)
parser.add_argument('--database', dest='database', help='The database that needs to be trained and tested.',
default='CVIQ', type=str)
parser.add_argument('--batch_size', dest='batch_size', help='Batch size.',
default=16, type=int)
parser.add_argument('--data_dir', dest='data_dir', help='Directory path for data.',
default='', type=str)
parser.add_argument('--filename_train', dest='filename_train', help='Training csv file containing relative paths for every example.',
default='', type=str)
parser.add_argument('--filename_test', dest='filename_test', help='Test csv file containing relative paths for every example.',
default='', type=str)
parser.add_argument('--cross_validation_index', dest='cross_validation_index', help='The index of cross validation.',
default='', type=int)
parser.add_argument('--snapshot', dest='snapshot', help='Path of model snapshot.',
default='', type=str)
args = parser.parse_args()
return args
if __name__ == '__main__':
args = parse_args()
cudnn.enabled = True
num_epochs = args.num_epochs
batch_size = args.batch_size
gpu = args.gpu_id
snapshot = args.snapshot
database = args.database
filename_train = args.filename_train
filename_test = args.filename_test
lr = args.lr
cross_validation_index = args.cross_validation_index
torch.cuda.set_device(gpu)
if not os.path.exists(os.path.join(snapshot, database, str(cross_validation_index))):
os.makedirs(os.path.join(snapshot, database, str(cross_validation_index)))
# load the network
model = multi_channel_resnet34_hyper.resnet34(pretrained = True)
transformations = transforms.Compose([transforms.Resize(224),transforms.ToTensor(),\
transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])])
train_dataset = dataset.Dataset(args.data_dir, filename_train, transformations)
train_loader = torch.utils.data.DataLoader(dataset=train_dataset, batch_size=batch_size, shuffle=True, num_workers=8)
test_dataset = dataset.Dataset(args.data_dir, filename_test, transformations)
test_loader = torch.utils.data.DataLoader(dataset=test_dataset, batch_size=1, shuffle=False, num_workers=8)
model.cuda()
criterion = nn.MSELoss().cuda()
# regression loss coefficient
optimizer = torch.optim.RMSprop(model.parameters(),lr=lr,alpha=0.9)
print("Ready to train network")
best_val_criterion = -1 # SROCC min
best_val = []
for epoch in range(num_epochs):
model.train()
epoch_start_time = time.time()
session_start_time = time.time()
batch_losses = []
batch_losses_each_disp = []
for i, (image_BA, image_BO, image_F, image_L, image_R, image_T, mos) in enumerate(train_loader):
image_BA = image_BA.cuda()
image_BO = image_BO.cuda()
image_F = image_F.cuda()
image_L = image_L.cuda()
image_R = image_R.cuda()
image_T = image_T.cuda()
mos = mos[:,np.newaxis]
mos = mos.cuda()
# Forward pass
mos_predict = model(image_BA,image_BO,image_F,image_L,image_R,image_T)
# MSE loss
loss = criterion(mos_predict,mos)
batch_losses.append(loss.item())
batch_losses_each_disp.append(loss.item())
optimizer.zero_grad() # clear gradients for next train
torch.autograd.backward(loss)
optimizer.step()
if (i+1) % 100 == 0:
session_end_time = time.time()
avg_loss_epoch = sum(batch_losses_each_disp) / 100
print('Epoch [%d/%d], Iter [%d/%d] Losses: %.4f CostTime: %.4f'
%(epoch+1, num_epochs, i+1, len(train_dataset)//batch_size, avg_loss_epoch, session_end_time - session_start_time))
session_start_time = time.time()
batch_losses_each_disp = []
avg_loss = sum(batch_losses) / (len(train_dataset) // batch_size)
print('Epoch %d averaged training loss: %.4f' % (epoch + 1, avg_loss))
# do validation after each epoch
with torch.no_grad():
model.eval()
label=np.zeros([len(test_dataset)])
y_output=np.zeros([len(test_dataset)])
for i, (image_BA, image_BO, image_F, image_L, image_R, image_T, mos) in enumerate(test_loader):
image_BA = image_BA.cuda()
image_BO = image_BO.cuda()
image_F = image_F.cuda()
image_L = image_L.cuda()
image_R = image_R.cuda()
image_T = image_T.cuda()
mos = mos.cuda()
label[i] = mos.item()
mos_predict = model(image_BA,image_BO,image_F,image_L,image_R,image_T)
y_output[i] = mos_predict.item()
label = np.array(label)
label = label.reshape(int(len(test_dataset)/180), 180)
label = np.mean(label, axis=1)
y_output = np.array(y_output)
y_output = y_output.reshape(int(len(test_dataset)/180), 180)
y_output = np.mean(y_output, axis=1)
y_output_logistic = fit_function(label, y_output)
val_PLCC = stats.pearsonr(y_output_logistic, label)[0]
val_SRCC = stats.spearmanr(y_output, label)[0]
val_KRCC = stats.stats.kendalltau(y_output, label)[0]
val_RMSE = np.sqrt(((y_output_logistic-label) ** 2).mean())
print('Epoch {} completed. SRCC: {:.4f}, KRCC: {:.4f}, PLCC: {:.4f}, and RMSE: {:.4f}'.format(epoch + 1, \
val_SRCC, val_KRCC, val_PLCC, val_RMSE))
if val_SRCC > best_val_criterion:
print("Update best model using best_val_criterion in epoch {}".format(epoch + 1))
best_val_criterion = val_SRCC
best_val = [val_SRCC, val_KRCC, val_PLCC, val_RMSE]
print('Saving model...')
torch.save(model.state_dict(), os.path.join(snapshot, database, str(cross_validation_index), database + '_epoch_'+ str(epoch+1) + '.pkl'))
print('Training completed.')
print('The best training result SRCC: {:.4f}, KRCC: {:.4f}, PLCC: {:.4f}, and RMSE: {:.4f}'.format( \
best_val[0], best_val[1], best_val[2], best_val[3]))