-
Notifications
You must be signed in to change notification settings - Fork 7
/
main.py
219 lines (182 loc) · 9.38 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
import torch
import torch.nn as nn
import torch.optim as optim
import argparse
import logging
import torchvision.transforms as transforms
import os
from util import *
import torch.cuda as cuda
import pickle
from torch.utils.data import Dataset, DataLoader
parser = argparse.ArgumentParser(description='Code for VISPE')
parser.add_argument('-e', '--epochs', action='store', default=300, type=int, help='epochs (default: 300)')
parser.add_argument('--batchSize', action='store', default=32, type=int, help='batch size (default: 32)')
parser.add_argument('--lr', '--learning-rate', action='store', default=0.001, type=float, help='learning rate (default: 0.001)')
parser.add_argument('--m', '--momentum', action='store', default=0.9, type=float, help='momentum (default: 0.9)')
parser.add_argument('--w', '--weight-decay', action='store', default=0, type=float, help='regularization weight decay (default: 0.0)')
parser.add_argument('--evaluate', action='store_true', default=False, help='Switch to evaluate mode (default: False)')
parser.add_argument('--gpu_num', type=int , default=0, help='gpu_num (default: 0)')
parser.add_argument('--load_pretrain', action='store_true', default=False, help='Flag to load pretrain model (default: False)')
parser.add_argument("--net", default='vgg16', const='vgg16',nargs='?', choices=['vgg16'], help="net model(default:vgg16)")
parser.add_argument("--dataset", default='modelnet40', const='modelnet40',nargs='?', choices=['modelnet40'], help="Dataset (default:modelnet40)")
parser.add_argument('--lamda', action='store', default=0.05, type=float, help='lamda (default: 0.05)')
parser.add_argument('--alpha', action='store', default=5, type=float, help='alpha (default: 5)')
parser.add_argument('--trial', action='store', default=1, type=int, help='trial (default: 1)')
arg = parser.parse_args()
def VISPE(model, x1, x2):
criterion_KL = nn.KLDivLoss(reduction='batchmean')
x1_feat = model(x1)
x2_feat = model(x2)
# prototype set 1
x1_x2_mat = torch.exp(torch.matmul(x1_feat,x2_feat.t())/arg.lamda)
denominator1 = torch.sum(x1_x2_mat, dim = 1).view(x1.shape[0],1)
prob1 = x1_x2_mat/denominator1
prob1_diag = torch.diag(prob1)
# prototype set 2
x2_x2_mat = torch.exp(torch.matmul(x2_feat,x2_feat.t())/arg.lamda)
I = torch.eye(x1.shape[0]).cuda()
x2_x2_mat = I*torch.diag(x1_x2_mat) + (1-I)*x2_x2_mat
denominator2 = torch.sum(x2_x2_mat, dim = 1).view(x1.shape[0],1)
prob2 = x2_x2_mat/denominator2
prob2_diag = torch.diag(prob2)
# KL divergence
loss_kl = criterion_KL(torch.log(prob1), prob2)
# cross entropy
loss_ce = -torch.mean(torch.log(prob1_diag)+torch.log(prob2_diag))
# Eq 8
loss = loss_ce + arg.alpha *loss_kl
return loss
def remove_duplicate_object(x1, x2, obj_IDs):
obj_set = set()
obj_idx = []
for i, oid in enumerate(obj_IDs):
if oid not in obj_set:
obj_idx.append(i)
obj_set.add(oid)
return x1[obj_idx], x2[obj_idx]
def main():
# create model directory to store/load old model
if not os.path.exists('model'):
os.makedirs('model')
if not os.path.exists('log'):
os.makedirs('log')
# Logger Setting
logger = logging.getLogger('netlog')
logger.setLevel(logging.INFO)
if arg.load_pretrain:
ch = logging.FileHandler('log/example.log')
else:
ch = logging.FileHandler('log/logfile_'+ arg.dataset + '_' + str(arg.trial) + '.log')
ch.setLevel(logging.INFO)
formatter = logging.Formatter('%(asctime)s - %(name)s - %(levelname)s - %(message)s')
ch.setFormatter(formatter)
logger.addHandler(ch)
logger.info("================================================")
logger.info("Learning Rate: {}".format(arg.lr))
logger.info("Momentum: {}".format(arg.m))
logger.info("Regularization Weight Decay: {}".format(arg.w))
logger.info("Classifier: "+arg.net)
logger.info("Dataset: "+arg.dataset)
logger.info("Nbr of Epochs: {}".format(arg.epochs))
logger.info("Lamda: {}".format(arg.lamda))
logger.info("Alpha: {}".format(arg.alpha))
logger.info("================================================")
# Batch size setting
batch_size = arg.batchSize
# load the data
data_transforms = {
'train': transforms.Compose([
transforms.Resize(256),
transforms.CenterCrop(224),
transforms.RandomHorizontalFlip(),
transforms.ToTensor(),
transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
]),
'val': transforms.Compose([
transforms.Resize(256),
transforms.CenterCrop(224),
transforms.ToTensor(),
transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
]),
'test': transforms.Compose([
transforms.Resize(256),
transforms.CenterCrop(224),
transforms.ToTensor(),
transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
]),
}
torch.cuda.set_device(arg.gpu_num)
# dataset directory
if arg.dataset == 'modelnet40':
pickle_filename = './dataset/modelnet/seen.pickle'
unseen_pickle_filename = './dataset/modelnet/unseen.pickle'
k = 960
seen_train, seen_test = load_data(pickle_filename)
unseen_train, unseen_test = load_data(unseen_pickle_filename)
dataset = {}
dataloader = {}
if not arg.evaluate:
dataset['train'] = mvDataset(seen_train, train=True, transform =data_transforms['train'])
dataloader['train'] = DataLoader(dataset['train'], batch_size=arg.batchSize, shuffle=True, num_workers=4)
dataset['seen_train_knn'] = mvDataset(seen_train, train=False, transform =data_transforms['test'])
dataloader['seen_train_knn'] = DataLoader(dataset['seen_train_knn'], batch_size=arg.batchSize, shuffle=False, num_workers=4)
dataset['seen_test_knn'] = mvDataset(seen_test, train=False, transform =data_transforms['test'])
dataloader['seen_test_knn'] = DataLoader(dataset['seen_test_knn'], batch_size=arg.batchSize, shuffle=False, num_workers=4)
dataset['unseen_train_knn'] = mvDataset(unseen_train, train=False, transform =data_transforms['test'])
dataloader['unseen_train_knn'] = DataLoader(dataset['unseen_train_knn'], batch_size=arg.batchSize, shuffle=False, num_workers=4)
dataset['unseen_test_knn'] = mvDataset(unseen_test, train=False, transform =data_transforms['test'])
dataloader['unseen_test_knn'] = DataLoader(dataset['unseen_test_knn'], batch_size=arg.batchSize, shuffle=False, num_workers=4)
if arg.net == 'vgg16':
model = vgg16()
optimizer = torch.optim.SGD(model.parameters(), lr=arg.lr, weight_decay=arg.w)
model.cuda()
model_path = 'model/model_'+ arg.dataset + '_' + str(arg.trial) +'.pt'
# training
print("Start Training")
logger.info("Start Training")
epochs = arg.epochs if not arg.evaluate else 0
min_acc = 0.0
for epoch in range(epochs):
model.train()
for batch_idx, (x1, x2, obj_IDs) in enumerate(dataloader['train']):
optimizer.zero_grad()
x1, x2 = remove_duplicate_object(x1, x2, obj_IDs)
loss = VISPE(model, x1.cuda(), x2.cuda())
loss.backward()
optimizer.step()
if batch_idx%50==0:
print('==>>> epoch:{}, batch index: {}, loss:{}'.format(epoch, batch_idx, loss.cpu().detach().numpy()))
logger.info('==>>> epoch:{}, batch index: {}, loss:{}'.format(epoch,batch_idx, loss.cpu().detach().numpy()))
# Validation (always save the best model)
print("Start Validation")
logger.info("Start Validation")
model.eval()
seen_acc = kNN(0, model, dataloader['seen_train_knn'], dataloader['seen_test_knn'], k, len(dataset['seen_train_knn']), low_dim = 4096)
unseen_acc = kNN(0, model, dataloader['unseen_train_knn'], dataloader['unseen_test_knn'], k, len(dataset['unseen_train_knn']), low_dim = 4096)
if seen_acc >= min_acc:
min_acc = seen_acc
torch.save(model.state_dict(), model_path)
print('==>>>test seen_acc:{} unseen_acc:{}'.format(seen_acc, unseen_acc))
logger.info('==>>>test seen_acc:{} unseen_acc:{}'.format(seen_acc, unseen_acc))
if arg.load_pretrain:
if os.path.isfile('model/example.pt'):
print("Loading pretrained model")
model.load_state_dict(torch.load('model/example.pt', map_location=lambda storage, loc: storage))
else:
print("No model")
return
else:
if os.path.isfile(model_path):
print("Loading model")
model.load_state_dict(torch.load(model_path, map_location=lambda storage, loc: storage))
else:
print("No model")
return
model.eval()
seen_acc = kNN(0, model, dataloader['seen_train_knn'], dataloader['seen_test_knn'], k, len(dataset['seen_train_knn']), low_dim = 4096)
unseen_acc = kNN(0, model, dataloader['unseen_train_knn'], dataloader['unseen_test_knn'], k, len(dataset['unseen_train_knn']), low_dim = 4096)
print('==>>>test seen_acc:{} unseen_acc:{}'.format(seen_acc, unseen_acc))
logger.info('==>>>test seen_acc:{} unseen_acc:{}'.format(seen_acc, unseen_acc))
if __name__ == "__main__":
main()