-
Notifications
You must be signed in to change notification settings - Fork 18
/
Copy pathSP.py
208 lines (197 loc) · 10 KB
/
SP.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
import torch
import torch.nn as nn
from torch.utils.data import Dataset, DataLoader
import torch.utils.model_zoo as model_zoo
import os
import numpy as np
from skimage import io
from scipy import ndimage
import time
from tqdm import tqdm
from floss import floss
from data.STdatas import STDataset
from models.model_SP import model_SP
from utils import *
class SP():
def __init__(self, lr=1e-7, loss_save='loss_SP.png', save_name='best_fusion.pth.tar', save_path='save', loss_function='f',\
num_epoch=10, batch_size=10, device='0', resume=1, pretrained_spatial='save/04_spatial.pth.tar', \
pretrained_temporal='save/03_temporal.pth.tar', traindata=None, valdata=None):
self.lr = lr
self.loss_save = loss_save
self.save_name = save_name
self.save_path = save_path
if not os.path.exists(save_path):
os.makedirs(save_path)
self.loss_function = loss_function
self.num_epoch = num_epoch
self.batch_size = batch_size
self.device = torch.device('cuda:'+device)
self.pretrained_spatial = pretrained_spatial
self.pretrained_temporal = pretrained_temporal
self.STTrainLoader = DataLoader(dataset=traindata, batch_size=batch_size, shuffle=True, num_workers=1, pin_memory=True)
self.STValLoader = DataLoader(dataset=valdata, batch_size=batch_size, shuffle=False, num_workers=1, pin_memory=True)
# 2: resume from fusion
# 0: from vgg
# 1: resume from separately pretrained models
in_channels = 20
if resume == '2':
self.model = model_SP(make_layers(cfg['D'], 3), make_layers(cfg['D'], 20))
trained_model = os.path.join(save_path, save_name)
pretrained_dict = torch.load(trained_model)
self.epochnow = pretrained_dict['epoch']
pretrained_optimizer = pretrained_dict['optimizer']
pretrained_dict = pretrained_dict['state_dict']
self.model.to(self.device)
model_dict = self.model.state_dict()
model_dict.update(pretrained_dict)
self.model.load_state_dict(model_dict)
elif resume == '0':
self.epochnow = 0
self.model = model_SP(make_layers(cfg['D'], 3), make_layers(cfg['D'], 20))
pretrained_dict = model_zoo.load_url('https://download.pytorch.org/models/vgg16_bn-6c64b313.pth')
model_dict_s = self.model.features_s.state_dict()
model_dict_t = self.model.features_t.state_dict()
new_pretrained_dict = change_key_names(pretrained_dict, in_channels)
new_pretrained_dict = {k: v for k, v in new_pretrained_dict.items() if 'features' in k}
pretrained_dict = {k: v for k,v in pretrained_dict.items() if 'features' in k}
pretrained_dict2 = {}
new_pretrained_dict2 = {}
for k in pretrained_dict.keys():
pretrained_dict2[k[9:]] = pretrained_dict[k]
for k in new_pretrained_dict.keys():
new_pretrained_dict2[k[9:]] = new_pretrained_dict[k]
new_pretrained_dict = {k: v for k, v in new_pretrained_dict2.items() if k in model_dict_t}
pretrained_dict = {k: v for k,v in pretrained_dict2.items() if k in model_dict_s}
model_dict_s.update(pretrained_dict)
model_dict_t.update(new_pretrained_dict)
self.model.features_s.load_state_dict(model_dict_s)
self.model.features_t.load_state_dict(model_dict_t)
self.model.to(self.device)
else:
self.epochnow = 0
self.model = model_SP(make_layers(cfg['D'], 3), make_layers(cfg['D'], 20))
pretrained_dict_s = torch.load(self.pretrained_spatial)
pretrained_dict_t = torch.load(self.pretrained_temporal)
model_dict_s = self.model.features_s.state_dict()
model_dict_t = self.model.features_t.state_dict()
pretrained_dict_s = pretrained_dict_s['state_dict']
pretrained_dict_t = pretrained_dict_t['state_dict']
pretrained_dict_s = {k: v for k,v in pretrained_dict_s.items() if 'features' in k}
pretrained_dict_t = {k: v for k,v in pretrained_dict_t.items() if 'features' in k}
new_pretrained_dict_t = {}
new_pretrained_dict_s = {}
for k in pretrained_dict_t.keys():
new_pretrained_dict_t[k[9:]] = pretrained_dict_t[k]
for k in pretrained_dict_s.keys():
new_pretrained_dict_s[k[9:]] = pretrained_dict_s[k]
new_pretrained_dict_s = {k: v for k,v in pretrained_dict_s.items() if k in model_dict_s}
new_pretrained_dict_t = {k: v for k,v in pretrained_dict_t.items() if k in model_dict_t}
model_dict_s.update(new_pretrained_dict_s)
model_dict_t.update(new_pretrained_dict_t)
self.model.features_s.load_state_dict(model_dict_s)
self.model.features_t.load_state_dict(model_dict_t)
self.model.to(self.device)
for params in self.model.features_s.parameters():
params.requires_grad = False
for params in self.model.features_t.parameters():
params.requires_grad = False
if loss_function != 'f':
self.criterion = torch.nn.BCELoss().to(self.device)
else:
self.criterion = floss().to(self.device)
# train params may be change according to resume state
if resume != '0':
self.optimizer = torch.optim.Adam([{'params': list(self.model.fusion.parameters())+list(self.model.bn.parameters())+list(self.model.decoder.parameters()),}
], lr=self.lr)
else:
self.optimizer = torch.optim.Adam(self.model.parameters(), lr = self.lr)
if resume == '2':
self.optimizer.load_state_dict(pretrained_optimizer)
print('SP module init done!')
def trainSP(self):
self.model.train()
batch_time = AverageMeter()
losses = AverageMeter()
end = time.time()
self.optimizer.zero_grad()
loss_mini_batch = 0.0
for i, sample in tqdm(enumerate(self.STTrainLoader)):
input_s = sample['image']
target = sample['gt']
input_t = sample['flow']
input_s = input_s.float().to(self.device)
input_t = input_t.float().to(self.device)
target = target.float().to(self.device)
output = self.model(input_s, input_t)
target = target.view(output.size())
loss = self.criterion(output, target)
loss_mini_batch += loss.item()
loss.backward()
self.optimizer.step()
self.optimizer.zero_grad()
batch_time.update(time.time() - end)
losses.update(loss.item(), input_s.size(0))
end = time.time()
loss_mini_batch = 0
if (i+1)%1000 == 0:
print('Epoch: [{0}][{1}/{2}]\t''Time {batch_time.val:.3f} ({batch_time.avg:.3f})\t''Loss {loss.val:.4f} ({loss.avg:.4f})\t'.format(
self.epochnow, i+1, len(self.STTrainLoader)+1, batch_time = batch_time, loss= losses))
return losses.avg
def testSP(self):
self.model.eval()
batch_time = AverageMeter()
losses = AverageMeter()
auc = AverageMeter()
aae = AverageMeter()
end = time.time()
with torch.no_grad():
for i, sample in tqdm(enumerate(self.STValLoader)):
input_s = sample['image']
target = sample['gt']
input_t = sample['flow']
input_s = input_s.float().to(self.device)
input_t = input_t.float().to(self.device)
target = target.float().to(self.device)
output = self.model(input_s, input_t)
target = target.view(output.size())
loss = self.criterion(output, target)
losses.update(loss.item(), input_s.size(0))
batch_time.update(time.time() - end)
end = time.time()
outim = output.cpu().data.numpy().squeeze()
targetim = target.cpu().data.numpy().squeeze()
aae1, auc1, _ = computeAAEAUC(outim, targetim)
auc.update(auc1)
aae.update(aae1)
'''
if i == 836: #inception of results, actually completely useless
outim = output.data.cpu().numpy()
outim = outim[0,:,:,:].squeeze()
io.imsave(os.path.join(self.save_path,'fusion_test_%05d.jpg'%i),outim)
if not os.path.exists(os.path.join(self.save_path,'targetfusion_%05d.jpg')):
targetim = target.data.cpu().numpy()
targetim = targetim[0,:,:,:].squeeze()
io.imsave(os.path.join(self.save_path,'targetfusion_%05d.jpg'%i), targetim)
'''
if (i+1) % 1000 == 0:
print('Test: [{0}/{1}]\t'
'Time {batch_time.val:.3f} ({batch_time.avg:.3f})\t'
'Loss {loss.val:.4f} ({loss.avg:.4f})\t'.format(i, len(self.STValLoader), batch_time=batch_time, loss=losses,))
print ('AUC: {0}\t AAE: {1}'.format(auc.avg, aae.avg))
return losses.avg, auc.avg, aae.avg
def train(self):
train_loss = []
val_loss = []
best_loss = 100
for epoch in range(self.epochnow, self.num_epoch):
self.epochnow = epoch
loss1 = self.trainSP()
train_loss.append(loss1)
loss1, auc1, aae1 = self.testSP()
val_loss.append(loss1)
plot_loss(train_loss, val_loss, os.path.join(self.save_path, self.loss_save))
checkpoint_name = self.save_name
if loss1 < best_loss:
best_loss = loss1
save_checkpoint({'epoch': epoch, 'arch': 'SP', 'state_dict': self.model.state_dict(),'optimizer':self.optimizer.state_dict(), 'auc': auc1, 'aae': aae1},
checkpoint_name, self.save_path)