forked from bruinxiong/SENet.mxnet
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathsymbol_se_resnext_w_d.py
163 lines (151 loc) · 9.87 KB
/
symbol_se_resnext_w_d.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
'''
Adapted from https://github.com/tornadomeet/ResNet/blob/master/symbol_resnet.py
Original author Wei Wu
Implemented the following paper:
Saining Xie, Ross Girshick, Piotr Dollar, Zhuowen Tu, Kaiming He. "Aggregated Residual Transformations for Deep Neural Network" CVPR 2017 https://arxiv.org/pdf/1611.05431v2.pdf
Jie Hu, Li Shen, Gang Sun. "Squeeze-and-Excitation Networks" https://arxiv.org/pdf/1709.01507v1.pdf
This modification version is based on ResNet v1
This modificaiton version adds dropout layer followed by last pooling layer.
Modified by Lin Xiong Feb-11, 2017
Updated by Lin Xiong Jul-21, 2017
Added Squeeze-and-Excitation block by Lin Xiong Sep-13, 2017
'''
import mxnet as mx
def residual_unit(data, num_filter, ratio, stride, dim_match, name, num_group, bottle_neck=True, bn_mom=0.9, workspace=256, memonger=False):
"""Return ResNext Unit symbol for building ResNext
Parameters
----------
data : str
Input data
num_filter : int
Number of output channels
stride : tupe
Stride used in convolution
dim_match : Boolen
True means channel number between input and output is the same, otherwise means differ
name : str
Base name of the operators
bottle_neck : Boolen
Whether or not to adopt bottle_neck trick as did in ResNet
num_group : int
Number of convolution groupes
bn_mom : float
Momentum of batch normalization
workspace : int
Workspace used in convolution operator
"""
if bottle_neck:
# the same as https://github.com/facebook/fb.resnet.torch#notes, a bit difference with origin paper
conv1 = mx.sym.Convolution(data=data, num_filter=int(num_filter*0.5), kernel=(1,1), stride=(1,1), pad=(0,0),
no_bias=True, workspace=workspace, name=name + '_conv1')
bn1 = mx.sym.BatchNorm(data=conv1, fix_gamma=False, eps=2e-5, momentum=bn_mom, name=name + '_bn1')
act1 = mx.sym.Activation(data=bn1, act_type='relu', name=name + '_relu1')
conv2 = mx.sym.Convolution(data=act1, num_filter=int(num_filter*0.5), num_group=num_group, kernel=(3,3), stride=stride, pad=(1,1),
no_bias=True, workspace=workspace, name=name + '_conv2')
bn2 = mx.sym.BatchNorm(data=conv2, fix_gamma=False, eps=2e-5, momentum=bn_mom, name=name + '_bn2')
act2 = mx.sym.Activation(data=bn2, act_type='relu', name=name + '_relu2')
conv3 = mx.sym.Convolution(data=act2, num_filter=num_filter, kernel=(1,1), stride=(1,1), pad=(0,0), no_bias=True,
workspace=workspace, name=name + '_conv3')
bn3 = mx.sym.BatchNorm(data=conv3, fix_gamma=False, eps=2e-5, momentum=bn_mom, name=name + '_bn3')
squeeze = mx.sym.Pooling(data=bn3, global_pool=True, kernel=(7, 7), pool_type='avg', name=name + '_squeeze')
squeeze = mx.symbol.Flatten(data=squeeze, name=name + '_flatten')
excitation = mx.symbol.FullyConnected(data=squeeze, num_hidden=int(num_filter*ratio), name=name + '_excitation1')
excitation = mx.sym.Activation(data=excitation, act_type='relu', name=name + '_excitation1_relu')
excitation = mx.symbol.FullyConnected(data=excitation, num_hidden=num_filter, name=name + '_excitation2')
excitation = mx.sym.Activation(data=excitation, act_type='sigmoid', name=name + '_excitation2_sigmoid')
bn3 = mx.symbol.broadcast_mul(bn3, mx.symbol.reshape(data=excitation, shape=(-1, num_filter, 1, 1)))
if dim_match:
shortcut = data
else:
shortcut_conv = mx.sym.Convolution(data=data, num_filter=num_filter, kernel=(1,1), stride=stride, no_bias=True,
workspace=workspace, name=name+'_sc')
shortcut = mx.sym.BatchNorm(data=shortcut_conv, fix_gamma=False, eps=2e-5, momentum=bn_mom, name=name + '_sc_bn')
if memonger:
shortcut._set_attr(mirror_stage='True')
eltwise = bn3 + shortcut
return mx.sym.Activation(data=eltwise, act_type='relu', name=name + '_relu')
else:
conv1 = mx.sym.Convolution(data=data, num_filter=num_filter, kernel=(3,3), stride=stride, pad=(1,1),
no_bias=True, workspace=workspace, name=name + '_conv1')
bn1 = mx.sym.BatchNorm(data=conv1, fix_gamma=False, momentum=bn_mom, eps=2e-5, name=name + '_bn1')
act1 = mx.sym.Activation(data=bn1, act_type='relu', name=name + '_relu1')
conv2 = mx.sym.Convolution(data=act1, num_filter=num_filter, kernel=(3,3), stride=(1,1), pad=(1,1),
no_bias=True, workspace=workspace, name=name + '_conv2')
bn2 = mx.sym.BatchNorm(data=conv2, fix_gamma=False, momentum=bn_mom, eps=2e-5, name=name + '_bn2')
squeeze = mx.sym.Pooling(data=bn2, global_pool=True, kernel=(7, 7), pool_type='avg', name=name + '_squeeze')
squeeze = mx.symbol.Flatten(data=squeeze, name=name + '_flatten')
excitation = mx.symbol.FullyConnected(data=squeeze, num_hidden=int(num_filter*ratio), name=name + '_excitation1')
excitation = mx.sym.Activation(data=excitation, act_type='relu', name=name + '_excitation1_relu')
excitation = mx.symbol.FullyConnected(data=excitation, num_hidden=num_filter, name=name + '_excitation2')
excitation = mx.sym.Activation(data=excitation, act_type='sigmoid', name=name + '_excitation2_sigmoid')
bn2 = mx.symbol.broadcast_mul(bn2, mx.symbol.reshape(data=excitation, shape=(-1, num_filter, 1, 1)))
if dim_match:
shortcut = data
else:
shortcut_conv = mx.sym.Convolution(data=data, num_filter=num_filter, kernel=(1,1), stride=stride, no_bias=True,
workspace=workspace, name=name+'_sc')
shortcut = mx.sym.BatchNorm(data=shortcut_conv, fix_gamma=False, eps=2e-5, momentum=bn_mom, name=name + '_sc_bn')
if memonger:
shortcut._set_attr(mirror_stage='True')
eltwise = bn2 + shortcut
return mx.sym.Activation(data=eltwise, act_type='relu', name=name + '_relu')
def resnext(units, num_stage, filter_list, ratio_list, num_class, num_group, data_type, drop_out, bottle_neck=True, bn_mom=0.9, workspace=256, memonger=False):
"""Return ResNeXt symbol of
Parameters
----------
units : list
Number of units in each stage
num_stages : int
Number of stage
filter_list : list
Channel size of each stage
num_class : int
Ouput size of symbol
num_groupes: int
Number of convolution groups
drop_out : float
Probability of an element to be zeroed. Default = 0.0
data_type : str
Dataset type, only cifar10, imagenet and vggface supports
workspace : int
Workspace used in convolution operator
"""
num_unit = len(units)
assert(num_unit == num_stage)
data = mx.sym.Variable(name='data')
data = mx.sym.BatchNorm(data=data, fix_gamma=True, eps=2e-5, momentum=bn_mom, name='bn_data')
if data_type == 'cifar10':
body = mx.sym.Convolution(data=data, num_filter=filter_list[0], kernel=(3, 3), stride=(1,1), pad=(1, 1),
no_bias=True, name="conv0", workspace=workspace)
elif data_type == 'imagenet':
body = mx.sym.Convolution(data=data, num_filter=filter_list[0], kernel=(7, 7), stride=(2,2), pad=(3, 3),
no_bias=True, name="conv0", workspace=workspace)
body = mx.sym.BatchNorm(data=body, fix_gamma=False, eps=2e-5, momentum=bn_mom, name='bn0')
body = mx.sym.Activation(data=body, act_type='relu', name='relu0')
body = mx.symbol.Pooling(data=body, kernel=(3, 3), stride=(2,2), pad=(1,1), pool_type='max')
elif data_type == 'vggface':
body = mx.sym.Convolution(data=data, num_filter=filter_list[0], kernel=(7, 7), stride=(2,2), pad=(3, 3),
no_bias=True, name="conv0", workspace=workspace)
body = mx.sym.BatchNorm(data=body, fix_gamma=False, eps=2e-5, momentum=bn_mom, name='bn0')
body = mx.sym.Activation(data=body, act_type='relu', name='relu0')
body = mx.symbol.Pooling(data=body, kernel=(3, 3), stride=(2,2), pad=(1,1), pool_type='max')
elif data_type == 'msface':
body = mx.sym.Convolution(data=data, num_filter=filter_list[0], kernel=(7, 7), stride=(2,2), pad=(3, 3),
no_bias=True, name="conv0", workspace=workspace)
body = mx.sym.BatchNorm(data=body, fix_gamma=False, eps=2e-5, momentum=bn_mom, name='bn0')
body = mx.sym.Activation(data=body, act_type='relu', name='relu0')
body = mx.symbol.Pooling(data=body, kernel=(3, 3), stride=(2,2), pad=(1,1), pool_type='max')
else:
raise ValueError("do not support {} yet".format(data_type))
for i in range(num_stage):
body = residual_unit(body, filter_list[i+1], ratio_list[2], (1 if i==0 else 2, 1 if i==0 else 2), False,
name='stage%d_unit%d' % (i + 1, 1), num_group=num_group, bottle_neck=bottle_neck,
bn_mom=bn_mom, workspace=workspace, memonger=memonger)
for j in range(units[i]-1):
body = residual_unit(body, filter_list[i+1], ratio_list[2], (1,1), True, name='stage%d_unit%d' % (i + 1, j + 2),
num_group=num_group, bottle_neck=bottle_neck, bn_mom=bn_mom, workspace=workspace, memonger=memonger)
pool1 = mx.symbol.Pooling(data=body, global_pool=True, kernel=(7, 7), pool_type='avg', name='pool1')
flat = mx.symbol.Flatten(data=pool1)
drop1= mx.symbol.Dropout(data=flat, p=drop_out, name='dp1')
fc1 = mx.symbol.FullyConnected(data=drop1, num_hidden=num_class, name='fc1')
return mx.symbol.SoftmaxOutput(data=fc1, name='softmax')