-
Notifications
You must be signed in to change notification settings - Fork 3
/
model_gao.py
145 lines (108 loc) · 4.43 KB
/
model_gao.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
import torch
import torch.nn as nn
import torch.nn.functional as F
class ReLU20(nn.Hardtanh):
def __init__(self, inplace=False):
super(ReLU20, self).__init__(0, 20, inplace)
def extra_repr(self):
inplace_str = 'inplace' if self.inplace else ''
return inplace_str
class gaoModel(nn.Module):
def __init__(self,embedding_size,num_classes,input_dim = 64,hidden_unit=[512,512,512,512,512],hidden_use_p=[(40, 5,1), (1, 3,2), (1, 3,4),(1,1,1), (1,1,1)],bottle=[512]):
super(gaoModel, self).__init__()
self.embedding_size = embedding_size
self.non_linear = ReLU20()
self.layer = []
for i,(unit,use_p) in enumerate(zip(hidden_unit,hidden_use_p)):
concat_layer = conv_frame(input_dim, unit, use_p, non_linear=self.non_linear)
input_dim = unit
self.layer.append(concat_layer)
setattr(self, 'cocat_layer{}'.format(i), concat_layer)
self.M_base = torch.zeros(input_dim, input_dim).cuda()
input_dim = input_dim * input_dim
self.bottleneck_layer = []
for i,output in enumerate(bottle):
concat_layer = bottleneck(input_dim, output ,non_linear=self.non_linear)
input_dim = output
self.bottleneck_layer.append(concat_layer)
setattr(self, 'bottle_layer{}'.format(i), concat_layer)
self.bottle_neck = nn.Linear(input_dim, self.embedding_size)
self.output_layer = nn.Linear(self.embedding_size, num_classes)
def l2_norm(self,input):
input_size = input.size()
buffer = torch.pow(input, 2)
normp = torch.sum(buffer, 1).add_(1e-10)
norm = torch.sqrt(normp)
_output = torch.div(input, norm.view(-1, 1).expand_as(input))
output = _output.view(input_size)
return output
def forward(self,input_x):
#input data dim. is batch X 1 X feature_size X time
if not input_x.shape[1] == 1:
exit(1)
x= input_x.view((input_x.shape[0],input_x.shape[2],input_x.shape[3]),-1)
x = x.transpose(1, 2)
for layer in self.layer[:-1]:
x = layer(x)
xb = self.layer[-1](x)
stack = []
for b_idx in range(xb.shape[0]):
M = self.M_base.clone()
for t_idx in range(xb.shape[2]):
M = torch.addr(M,x[b_idx,:,t_idx],xb[b_idx,:,t_idx])
stack.append(M.view(-1))
x = torch.stack(stack)
for layer in self.bottleneck_layer:
x = layer(x)
x = self.bottle_neck(x)
self.features = self.l2_norm(x)
# Multiply by alpha = 10 as suggested in https://arxiv.org/pdf/1703.09507.pdf
alpha = 10
self.features = self.features * alpha
return self.features
def forward_classifier(self, x):
features = self.forward(x)
res = self.output_layer(features)
return res
class conv_frame(nn.Module):
def __init__(self, input_dim,output_dim, use_p,non_linear):
super(conv_frame,self).__init__()
self.input_dim = input_dim
self.output_dim = output_dim
self.use_p = use_p
self.kernel_width= use_p[1]
self.dilation = use_p[2]
if input_dim == use_p[0] or 1 == use_p[0]:
if self.kernel_width == 1:
self.flag = True
self.conv_h = nn.Linear(input_dim, output_dim)
else:
self.flag = False
self.conv_h = nn.Conv1d(input_dim, output_dim, self.kernel_width,dilation = self.dilation)
else:
"""this is not implement"""
self.conv_h_bn = nn.BatchNorm1d(self.output_dim)
self.non_linear = non_linear
def forward(self,input_x):
if self.flag:
input_x = input_x.transpose(1, 2)
x = self.conv_h(input_x)
x = x.transpose(1, 2)
else:
x = self.conv_h(input_x)
x = self.conv_h_bn(x)
x = self.non_linear(x)
return x
class bottleneck(nn.Module):
def __init__(self, input_dim,output_dim, non_linear):
super(bottleneck,self).__init__()
self.input_dim = input_dim
self.output_dim = output_dim
self.conv_h = nn.Linear(input_dim, output_dim)
self.conv_h_bn = nn.BatchNorm1d(self.output_dim)
self.non_linear = non_linear
def forward(self,input_x):
x = self.conv_h(input_x)
x = self.conv_h_bn(x)
x = self.non_linear(x)
return x