/
train_MMD.py
164 lines (145 loc) · 6.85 KB
/
train_MMD.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
import os
import numpy as np
import torch
import torch.optim as optim
from torch.nn import functional as F
from methods.backbone import model_dict
from data.datamgr import SetDataManager
from methods.matchingnet import MatchingNet
from methods.relationnet import RelationNet
from methods.protonet import ProtoNet
from methods.gnnnet import GnnNet
from methods.tpn import TPN
from options import parse_args
def guassian_kernel(source, target, kernel_mul, kernel_num, fix_sigma=None):
n_samples = int(source.size()[0])+int(target.size()[0]) # n + m
total = torch.cat([source, target], dim=0) # (n+m, d)
AB = torch.mm(total, total.transpose(0, 1)) # (n+m, n+m)
AA = (total * total).sum(dim=1, keepdim=True).expand_as(AB) # (n+m, n+m)
BB = (total * total).sum(dim=1).unsqueeze(0).expand_as(AB) # (n+m, n+m)
L2_distance = AA - 2.*AB + BB
if fix_sigma:
bandwidth = fix_sigma
else:
bandwidth = torch.sum(L2_distance.detach().data) / (n_samples**2-n_samples)
bandwidth /= kernel_mul**(kernel_num//2)
bandwidth_list = [bandwidth * (kernel_mul**i) for i in range(kernel_num)]
kernel_val = [torch.exp(-L2_distance / bandwidth_temp) for bandwidth_temp in bandwidth_list]
return sum(kernel_val)
def mmd(source, target, kernel_mul=2., kernel_num=5, fix_sigma=None):
batch_size = int(source.size()[0])
kernels = guassian_kernel(source, target, kernel_mul=kernel_mul, kernel_num=kernel_num, fix_sigma=fix_sigma)
XX = kernels[:batch_size, :batch_size]
YY = kernels[batch_size:, batch_size:]
XY = kernels[:batch_size, batch_size:]
YX = kernels[batch_size:, :batch_size]
loss = torch.mean(XX + YY - XY - YX)
return loss
def Max_phase(model, X_n):
X_n = X_n.cuda()
optimizer = optim.SGD([X_n.requires_grad_()], lr=params.max_lr)
model.eval()
init_features = None
for i in range(params.T_max):
optimizer.zero_grad()
last_features = model.feature(X_n.reshape(-1, *X_n.size()[2:])) # (105, 512)
if params.method in ['RelationNet', 'TPN']:
last_features = F.avg_pool2d(last_features, kernel_size=7).squeeze() # [105, 512]
if i == 0:
init_features = last_features.clone().detach() # (105, 512)
_, class_loss = model.set_forward_loss(X_n)
feature_loss = mmd(last_features, init_features)
adv_loss = params.lamb * feature_loss - class_loss
adv_loss.backward()
optimizer.step()
del last_features, class_loss, feature_loss, adv_loss
return X_n.detach()
def train(base_loader, val_loader, model, start_epoch, stop_epoch, params):
max_acc = 0.
optimizer = torch.optim.Adam(model.parameters())
print_freq = len(base_loader)//10
for epoch in range(start_epoch, stop_epoch):
avg_loss = 0.
for i, (x, _) in enumerate(base_loader): # (5, 21, 3, 224, 224)
x_hat = Max_phase(model, x) # (5, 21, 3, 224, 224)
model.train()
optimizer.zero_grad()
_, loss = model.set_forward_loss(x_hat)
loss.backward()
optimizer.step()
avg_loss = avg_loss + loss.item()
if (i + 1) % print_freq == 0:
print('Epoch {:d} | Batch {:d}/{:d} | Loss {:f}'.format(epoch, i+1, len(base_loader), avg_loss/float(i+1)))
model.eval()
with torch.no_grad():
acc = model.test_loop(val_loader)
if acc > max_acc:
print("Best model! save...")
max_acc = acc
outfile = os.path.join(params.checkpoint_dir, 'best_model.tar')
torch.save({'epoch':epoch, 'state':model.state_dict()}, outfile)
else:
print("GG! Best accuracy {:f}".format(max_acc))
if ((epoch+1) % params.save_freq == 0) or (epoch == stop_epoch-1):
outfile = os.path.join(params.checkpoint_dir, '{:d}.tar'.format(epoch))
torch.save({'epoch': epoch, 'state': model.state_dict()}, outfile)
return model
# --- main function ---
if __name__=='__main__':
# set numpy random seed
np.random.seed(10)
# parser argument
params = parse_args()
print('--- Training ---\n')
print(params)
# output and tensorboard dir
params.checkpoint_dir = '%s/checkpoints/%s'%(params.save_dir, params.name)
if not os.path.isdir(params.checkpoint_dir):
os.makedirs(params.checkpoint_dir)
# dataloader
print('\n--- Prepare dataloader ---')
print('\ttrain with single seen domain {}'.format(params.dataset))
print('\tval with single seen domain {}'.format(params.testset))
base_file = os.path.join(params.data_dir, params.dataset, 'base.json')
val_file = os.path.join(params.data_dir, params.testset, 'val.json')
# model
image_size = 224
n_query = max(1, int(16*params.test_n_way/params.train_n_way))
base_datamgr = SetDataManager(image_size, n_query=n_query, n_way=params.train_n_way, n_support=params.n_shot)
base_loader = base_datamgr.get_data_loader(base_file, aug=params.train_aug)
val_datamgr = SetDataManager(image_size, n_query=n_query, n_way=params.test_n_way, n_support=params.n_shot)
val_loader = val_datamgr.get_data_loader(val_file, aug=False)
if params.method == 'MatchingNet':
model = MatchingNet(model_dict[params.model], n_way=params.train_n_way, n_support=params.n_shot).cuda()
elif params.method == 'RelationNet':
model = RelationNet(model_dict[params.model], n_way=params.train_n_way, n_support=params.n_shot).cuda()
elif params.method == 'ProtoNet':
model = ProtoNet(model_dict[params.model], n_way=params.train_n_way, n_support=params.n_shot).cuda()
elif params.method == 'GNN':
model = GnnNet(model_dict[params.model], n_way=params.train_n_way, n_support=params.n_shot).cuda()
elif params.method == 'TPN':
model = TPN(model_dict[params.model], n_way=params.train_n_way, n_support=params.n_shot).cuda()
else:
print("Please specify the method!")
assert(False)
model.n_query = n_query
# load model
start_epoch = params.start_epoch
stop_epoch = params.stop_epoch
if params.resume_epoch > 0:
resume_file = os.path.join(params.checkpoint_dir, '{:d}.tar'.format(params.resume_epoch))
tmp = torch.load(resume_file)
start_epoch = tmp['epoch']+1
model.load_state_dict(tmp['state'])
print('\tResume the training weight at {} epoch.'.format(start_epoch))
else:
path = '%s/checkpoints/%s/399.tar' % (params.save_dir, params.resume_dir)
state = torch.load(path)['state']
model_params = model.state_dict()
pretrained_dict = {k: v for k, v in state.items() if k in model_params}
print(pretrained_dict.keys())
model_params.update(pretrained_dict)
model.load_state_dict(model_params)
# training
print('\n--- start the training ---')
model = train(base_loader, val_loader, model, start_epoch, stop_epoch, params)