forked from apache/mxnet
-
Notifications
You must be signed in to change notification settings - Fork 0
/
symbol_resnet-v2.R
145 lines (131 loc) · 6.61 KB
/
symbol_resnet-v2.R
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
###
# Reproducing parper:
# Kaiming He, Xiangyu Zhang, Shaoqing Ren, Jian Sun. "Identity Mappings in Deep Residual Networks"
###
library(mxnet)
residual_unit <- function(data, num_filter, stride, dim_match, name, bottle_neck=TRUE, bn_mom=0.9, workspace=512){
if(bottle_neck){
bn1 <- mx.symbol.BatchNorm(data=data, fix_gamma=FALSE, eps=2e-5,
momentum=bn_mom, name=paste0(name,'_bn1'))
act1 <- mx.symbol.Activation(data=bn1, act_type='relu',
name=paste0(name, '_relu1'))
conv1 <- mx.symbol.Convolution(data=act1, num_filter=as.integer(num_filter*0.25),
kernel=c(1,1), stride=c(1,1), pad=c(0,0),
no_bias=TRUE, workspace=workspace,
name=paste0(name,'_conv1'))
bn2 <- mx.symbol.BatchNorm(data=conv1, fix_gamma=FALSE, eps=2e-5,
momentum=bn_mom, name=paste0(name, '_bn2'))
act2 <- mx.symbol.Activation(data=bn2, act_type='relu', name=paste0(name, '_relu2'))
conv2 <- mx.symbol.Convolution(data=act2, num_filter=as.integer(num_filter*0.25),
kernel=c(3,3), stride=stride, pad=c(1,1),
no_bias=TRUE, workspace=workspace,
name=paste0(name, '_conv2'))
bn3 <- mx.symbol.BatchNorm(data=conv2, fix_gamma=FALSE, eps=2e-5,
momentum=bn_mom, name=paste0(name, '_bn3'))
act3 <- mx.symbol.Activation(data=bn3, act_type='relu', name=paste0(name,'_relu3'))
conv3 <- mx.symbol.Convolution(data=act3, num_filter=num_filter, kernel=c(1,1),
stride=c(1,1), pad=c(0,0), no_bias=TRUE,
workspace=workspace, name=paste0(name, '_conv3'))
if (dim_match){
shortcut <- data
} else{
shortcut <- mx.symbol.Convolution(data=act1, num_filter=num_filter,
kernel=c(1,1), stride=stride, no_bias=TRUE,
workspace=workspace, name=paste0(name,'_sc'))
}
return (conv3 + shortcut)
} else{
bn1 <- mx.symbol.BatchNorm(data=data, fix_gamma=FALSE, momentum=bn_mom,
eps=2e-5, name=paste0(name,'_bn1'))
act1 <- mx.symbol.Activation(data=bn1, act_type='relu', name=paste0(name, '_relu1'))
conv1 <- mx.symbol.Convolution(data=act1, num_filter=num_filter, kernel=c(3,3),
stride=stride, pad=c(1,1), no_bias=TRUE,
workspace=workspace, name=paste0(name,'_conv1'))
bn2 <- mx.symbol.BatchNorm(data=conv1, fix_gamma=FALSE, momentum=bn_mom,
eps=2e-5, name=paste0(name, '_bn2'))
act2 <- mx.symbol.Activation(data=bn2, act_type='relu',
name=paste0(name, '_relu2'))
conv2 <- mx.symbol.Convolution(data=act2, num_filter=num_filter, kernel=c(3,3),
stride=c(1,1), pad=c(1,1), no_bias=TRUE,
workspace=workspace, name=paste0(name, '_conv2'))
if (dim_match){
shortcut = data
} else {
shortcut <- mx.symbol.Convolution(data=act1, num_filter=num_filter, kernel=c(1,1),
stride=stride, no_bias=TRUE,
workspace=workspace, name=paste0(name,'_sc'))
}
return (conv2 + shortcut)
}
}
resnet <- function(units, num_stage, filter_list, num_class, bottle_neck=TRUE,
bn_mom=0.9, workspace=512){
num_unit <- length(units)
if(num_unit != num_stage) stop("Number of units different from num_stage")
data <- mx.symbol.Variable(name='data')
data <- mx.symbol.BatchNorm(data=data, fix_gamma=TRUE, eps=2e-5, momentum=bn_mom,
name='bn_data')
body <- mx.symbol.Convolution(data=data, num_filter=filter_list[1], kernel=c(7, 7),
stride=c(2,2), pad=c(3, 3),
no_bias=TRUE, name="conv0", workspace=workspace)
body <- mx.symbol.BatchNorm(data=body, fix_gamma=FALSE, eps=2e-5,
momentum=bn_mom, name='bn0')
body <- mx.symbol.Activation(data=body, act_type='relu', name='relu0')
body <- mx.symbol.Pooling(data=body, kernel=c(3, 3), stride=c(2,2),
pad=c(1,1), pool_type='max')
for(i in 1:num_stage){
if(i==1) stride <- c(1,1)
else stride <- c(2,2)
body <- residual_unit(body, filter_list[i+1], stride, FALSE,
name=paste0('stage', i, '_unit1') ,
bottle_neck=bottle_neck, workspace=workspace)
for(j in 1:(units[i]-1)){
body <- residual_unit(body, filter_list[i+1], c(1,1),
TRUE, name=paste0('stage',i, '_unit', j + 1),
bottle_neck=bottle_neck,
workspace=workspace)
}
}
bn1 <- mx.symbol.BatchNorm(data=body, fix_gamma=FALSE, eps=2e-5,
momentum=bn_mom, name='bn1')
relu1 <- mx.symbol.Activation(data=bn1, act_type='relu', name='relu1')
# Although kernel is not used here when global_pool=TRUE, we should put one
pool1 <- mx.symbol.Pooling(data=relu1, global_pool=TRUE, kernel=c(7, 7),
pool_type='avg', name='pool1')
flat <- mx.symbol.Flatten(data=pool1)
fc1 <- mx.symbol.FullyConnected(data=flat, num_hidden=num_class, name='fc1')
resnet <- mx.symbol.SoftmaxOutput(data=fc1, name='softmax')
return(resnet)
}
get_symbol <- function(num_class, depth=18){
if (depth == 18){
units <- c(2, 2, 2, 2)
} else if (depth == 34){
units = c(3, 4, 6, 3)
} else if (depth == 50){
units = c(3, 4, 6, 3)
} else if (depth == 101){
units = c(3, 4, 23, 3)
} else if (depth == 152){
units = c(3, 8, 36, 3)
} else if (depth == 200){
units = c(3, 24, 36, 3)
} else if (depth == 269){
units = c(3, 30, 48, 8)
} else{
stop(paste0("no experiments done on depth ", depth))
}
if (depth >=50){
filter_list <- c(64, 256, 512, 1024, 2048)
bottle_neck <- TRUE
} else{
filter_list <- c(64, 64, 128, 256, 512)
bottle_neck <- FALSE
}
bn_mom <- 0.9 #momentum of batch normalization
workspace <- 500
symbol <- resnet(units=units, num_stage=4, filter_list=filter_list,
num_class=num_class, bottle_neck=bottle_neck,
bn_mom=bn_mom, workspace=workspace)
return(symbol)
}