-
Notifications
You must be signed in to change notification settings - Fork 1
/
frmodels.py
206 lines (186 loc) · 7.91 KB
/
frmodels.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
import torch
import torch.nn as nn
from torch.nn import Linear, Conv2d, BatchNorm1d, BatchNorm2d, PReLU, ReLU, Sigmoid, Dropout, MaxPool2d, \
AdaptiveAvgPool2d, Sequential, Module
from collections import namedtuple
import numpy as np
import pdb
import torch.nn.functional as F
class L2pooling(nn.Module):
def __init__(self, filter_size=5, stride=2, channels=None, pad_off=0):
super(L2pooling, self).__init__()
self.padding = (filter_size - 2 )//2
self.stride = stride
self.channels = channels
a = np.hanning(filter_size)[1:-1]
# a = torch.hann_window(5,periodic=False)
g = torch.Tensor(a[:,None]*a[None,:])
g = g/torch.sum(g)
self.register_buffer('filter', g[None,None,:,:].repeat((self.channels,1,1,1)))
def forward(self, input):
input = input**2
out = F.conv2d(input, self.filter, stride=self.stride, padding=self.padding, groups=input.shape[1])
return (out+1e-12).sqrt()
class Flatten(Module):
'''
This method is to flatten the features
'''
def forward(self, input):
return input.view(input.size(0), -1)
def l2_norm(input, axis=1):
'''
This method is for l2 normalization
'''
norm = torch.norm(input, 2, axis, True)
output = torch.div(input, norm)
return output
class bottleneck_IR(Module):
def __init__(self, in_channel, depth, stride):
'''
This method is to initialize IR module
'''
super(bottleneck_IR, self).__init__()
if in_channel == depth:
self.shortcut_layer = MaxPool2d(1, stride)
# replace maxpooling to l2pooling
#self.shortcut_layer = L2pooling(channels=in_channel)
else:
self.shortcut_layer = Sequential(
Conv2d(in_channel, depth, (1, 1), stride, bias=False), BatchNorm2d(depth))
self.res_layer = Sequential(
BatchNorm2d(in_channel),
Conv2d(in_channel, depth, (3, 3), (1, 1), 1, bias=False),
PReLU(depth),
Conv2d(depth, depth, (3, 3), stride, 1, bias=False),
BatchNorm2d(depth))
def forward(self, x):
shortcut = self.shortcut_layer(x)
res = self.res_layer(x)
#print("res:", res.size())
#print("short:", shortcut.size())
return res + shortcut
class Bottleneck(namedtuple('Block', ['in_channel', 'depth', 'stride'])):
'''A named tuple describing a ResNet block.'''
def get_block(in_channel, depth, num_units, stride=2):
'''
This method is to obtain blocks
'''
return [Bottleneck(in_channel, depth, stride)] + [Bottleneck(depth, depth, 1) for i in range(num_units - 1)]
def get_blocks(num_layers):
'''
This method is to obtain blocks
'''
if num_layers == 50:
blocks = [
get_block(in_channel=64, depth=64, num_units=3),
get_block(in_channel=64, depth=128, num_units=4),
get_block(in_channel=128, depth=256, num_units=14),
get_block(in_channel=256, depth=512, num_units=3)
]
elif num_layers == 100:
blocks = [
get_block(in_channel=64, depth=64, num_units=3),
get_block(in_channel=64, depth=128, num_units=13),
get_block(in_channel=128, depth=256, num_units=30),
get_block(in_channel=256, depth=512, num_units=3)
]
elif num_layers == 152:
blocks = [
get_block(in_channel=64, depth=64, num_units=3),
get_block(in_channel=64, depth=128, num_units=8),
get_block(in_channel=128, depth=256, num_units=36),
get_block(in_channel=256, depth=512, num_units=3)
]
return blocks
class Backbone(Module):
def __init__(self, input_size, num_layers, mode='ir', use_type = "Rec"):
'''
This method is to initialize model
if use for quality network, select self.use_type == "Qua"
if use for recognition network, select self.use_type == "Rec"
'''
super(Backbone, self).__init__()
assert input_size[0] in [112, 224], "input_size should be [112, 112] or [224, 224]"
assert num_layers in [50, 100, 152], "num_layers should be 50, 100 or 152"
assert mode in ['ir', 'ir_se'], "mode should be ir or ir_se"
blocks = get_blocks(num_layers)
if mode == 'ir': unit_module = bottleneck_IR
self.input_layer = Sequential(Conv2d(3, 64, (3, 3), 1, 1, bias=False),
BatchNorm2d(64),
PReLU(64))
self.use_type = use_type
if input_size[0] == 112:
if use_type == "Qua":
self.output_layer = Sequential(BatchNorm2d(512),
Dropout(0.4),
Flatten(),
Linear(512 * 7 * 7, 512),
BatchNorm1d(512, affine=False)
)
else:
self.output_layer = Sequential(BatchNorm2d(512),
Dropout(0.4),
Flatten(),
Linear(512 * 14 * 14, 512),
BatchNorm1d(512, affine=False)
)
# if input_size[0] == 112:
# if use_type == "Qua":
# self.quality = Sequential(Flatten(),
# PReLU(512 * 7 * 7),
# Dropout(0.5, inplace=False),
# Linear(512 * 7 * 7, 1)
# )
# else:
# self.output_layer = Sequential(Flatten(),
# PReLU(512 * 7 * 7),
# Dropout(0.5, inplace=False),
# Linear(512 * 7 * 7, 512)
# )
#modify output_layer according to fuzhao's new sent:
modules = []
for block in blocks:
for bottleneck in block:
modules.append(
unit_module(bottleneck.in_channel,
bottleneck.depth,
bottleneck.stride))
self.body = Sequential(*modules)
self._initialize_weights()
def forward(self, x):
'''
This method is to model forward
'''
x = self.input_layer(x)
#print("size of inputlayer:", x.size)
x = self.body(x)
if self.use_type == "Qua":
x = self.output_layer(x)
else:
x = self.output_layer(x)
#print("outputlayersize:", x.size())
return x
def _initialize_weights(self):
'''
This method is to initialize model weights
'''
for m in self.modules():
if isinstance(m, nn.Conv2d):
nn.init.kaiming_normal_(m.weight, mode='fan_out', nonlinearity='relu')
if m.bias is not None:
m.bias.data.zero_()
elif isinstance(m, nn.BatchNorm2d):
m.weight.data.fill_(1)
m.bias.data.zero_()
elif isinstance(m, nn.Linear):
nn.init.kaiming_normal_(m.weight, mode='fan_out', nonlinearity='relu')
if m.bias is not None:
m.bias.data.zero_()
def R50(input_size, use_type = "Qua"):
'''
This method is to create ResNet50 backbone
if use for quality network, select self.use_type == "Qua"
if use for recognition network, select self.use_type == "Rec"
'''
model = Backbone(input_size, 50, 'ir', use_type = use_type)
return model