-
Notifications
You must be signed in to change notification settings - Fork 1.6k
/
torch_gen_test_data.lua
223 lines (185 loc) · 8.18 KB
/
torch_gen_test_data.lua
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
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
import 'nn'
import 'dpnn'
function fill_net(net, training)
if net.modules then
for i = 1, #net.modules do
fill_net(net.modules[i], training)
end
end
if net.weight then
net.weight = torch.rand(net.weight:size())
end
if net.bias then
net.bias = torch.rand(net.bias:size())
end
if (training == nil or not training) and net.train then
net.train = 0
end
end
function save(net, input, label, is_binary, training)
fill_net(net, training)
output = net:forward(input)
net:apply(function(module)
module.gradInput = nil
module.fgradInput = nil
module.output = nil
module.gradBias = nil
module.gradWeight = nil
module.input = nil
module.finput = nil
end)
local suffix = is_binary and '.dat' or '.txt'
local format = is_binary and 'binary' or 'ascii'
torch.save(label .. '_net' .. suffix, net, format)
torch.save(label .. '_input' .. suffix, input, format)
torch.save(label .. '_output' .. suffix, output, format)
return net
end
local net_simple = nn.Sequential()
net_simple:add(nn.ReLU())
net_simple:add(nn.SpatialConvolution(3,64, 11,7, 3,4, 3,2))
net_simple:add(nn.SpatialMaxPooling(4,5, 3,2, 1,2))
net_simple:add(nn.Sigmoid())
save(net_simple, torch.Tensor(2, 3, 25, 35), 'net_simple')
local net_pool_max = nn.Sequential()
net_pool_max:add(nn.SpatialMaxPooling(4,5, 3,2, 1,2):ceil())
local net = save(net_pool_max, torch.rand(2, 3, 50, 30), 'net_pool_max')
torch.save('net_pool_max_output_2.txt', net.modules[1].indices - 1, 'ascii')
local net_pool_ave = nn.Sequential()
net_pool_ave:add(nn.SpatialAveragePooling(4,5, 2,1, 1,2))
save(net_pool_ave, torch.rand(2, 3, 50, 30), 'net_pool_ave')
local net_conv = nn.Sequential()
net_conv:add(nn.SpatialConvolution(3,64, 11,7, 3,4, 3,2))
save(net_conv, torch.rand(1, 3, 50, 60), 'net_conv', true)
local net_reshape = nn.Sequential()
net_reshape:add(nn.Reshape(5, 4, 3, 2))
save(net_reshape, torch.rand(2, 3, 4, 5), 'net_reshape')
local net_reshape_batch = nn.Sequential()
net_reshape_batch:add(nn.Reshape(5, 4, 3, true))
save(net_reshape_batch, torch.rand(2, 3, 4, 5), 'net_reshape_batch')
local net_reshape_single_sample = nn.Sequential()
net_reshape_single_sample:add(nn.Reshape(3 * 4 * 5))
net_reshape_single_sample:add(nn.Linear(3 * 4 * 5, 10))
save(net_reshape_single_sample, torch.rand(1, 3, 4, 5), 'net_reshape_single_sample')
save(nn.Linear(7, 3), torch.rand(13, 7), 'net_linear_2d')
local net_reshape_channels = nn.Sequential()
net_reshape_channels:add(nn.Reshape(20))
save(net_reshape_channels, torch.rand(2, 1, 10, 2), 'net_reshape_channels', true)
local net_parallel = nn.Parallel(4, 2)
net_parallel:add(nn.Sigmoid())
net_parallel:add(nn.Tanh())
save(net_parallel, torch.rand(2, 6, 4, 2), 'net_parallel')
local net_concat = nn.Concat(2)
net_concat:add(nn.ReLU())
net_concat:add(nn.Tanh())
net_concat:add(nn.Sigmoid())
save(net_concat, torch.rand(2, 6, 4, 3) - 0.5, 'net_concat')
local net_deconv = nn.Sequential()
net_deconv:add(nn.SpatialFullConvolution(3, 9, 4, 5, 1, 2, 0, 1, 0, 1))
save(net_deconv, torch.rand(2, 3, 4, 3) - 0.5, 'net_deconv')
local net_batch_norm = nn.Sequential()
net_batch_norm:add(nn.SpatialBatchNormalization(4, 1e-3))
save(net_batch_norm, torch.rand(1, 4, 5, 6) - 0.5, 'net_batch_norm', true)
local net_batch_norm_train = nn.Sequential()
net_batch_norm_train:add(nn.SpatialBatchNormalization(4))
save(net_batch_norm_train, torch.randn(1, 4, 5, 6), 'net_batch_norm_train', true, --[[training]]true)
local net_prelu = nn.Sequential()
net_prelu:add(nn.PReLU(5))
save(net_prelu, torch.rand(1, 5, 40, 50) - 0.5, 'net_prelu')
local net_cadd_table = nn.Sequential()
local sum = nn.ConcatTable()
sum:add(nn.Identity()):add(nn.Identity())
net_cadd_table:add(sum):add(nn.CAddTable())
save(net_cadd_table, torch.rand(1, 5, 40, 50) - 0.5, 'net_cadd_table')
local net_softmax = nn.Sequential()
net_softmax:add(nn.SoftMax())
save(net_softmax, torch.rand(1, 5, 1, 1), 'net_softmax')
local net_softmax_spatial = nn.Sequential()
net_softmax_spatial:add(nn.SoftMax())
save(net_softmax_spatial, torch.rand(2, 5, 3, 4), 'net_softmax_spatial')
local net_logsoftmax = nn.Sequential()
net_logsoftmax:add(nn.LogSoftMax())
save(net_logsoftmax, torch.rand(1, 6, 4, 3), 'net_logsoftmax_spatial')
local net_logsoftmax_spatial = nn.Sequential()
net_logsoftmax_spatial:add(nn.SoftMax())
save(net_logsoftmax_spatial, torch.rand(1, 6, 4, 3), 'net_logsoftmax_spatial')
local net_lp_pooling_square = nn.Sequential()
net_lp_pooling_square:add(nn.SpatialLPPooling(-1, 2, 2,2, 2,2)) -- The first argument isn't used
net_lp_pooling_square:add(nn.Tanh())
save(net_lp_pooling_square, torch.rand(3, 7, 8, 10), 'net_lp_pooling_square', true)
local net_lp_pooling_power = nn.Sequential()
net_lp_pooling_power:add(nn.SpatialLPPooling(-1, 3, 3,3, 2,2)) -- The first argument isn't used
net_lp_pooling_power:add(nn.Sigmoid())
save(net_lp_pooling_power, torch.rand(3, 7, 6, 7), 'net_lp_pooling_power', true)
local net_conv_gemm_lrn = nn.Sequential()
net_conv_gemm_lrn:add(nn.SpatialConvolutionMM(4,7, 3,3, 1,1, 1,1))
net_conv_gemm_lrn:add(nn.SpatialCrossMapLRN(3))
save(net_conv_gemm_lrn, torch.rand(2, 4, 5, 6), 'net_conv_gemm_lrn', true)
local net_depth_concat = nn.DepthConcat(1);
net_depth_concat:add(nn.SpatialConvolutionMM(3, 4, 1, 1))
net_depth_concat:add(nn.SpatialConvolutionMM(3, 5, 3, 3))
net_depth_concat:add(nn.SpatialConvolutionMM(3, 2, 4, 4))
save(net_depth_concat, torch.rand(2, 3, 7, 7), 'net_depth_concat', true)
local net_inception_block = nn.Sequential()
net_inception_block:add(nn.Inception{
inputSize = 3, -- Number of input channels
kernelSize = {3},
kernelStride = {1},
outputSize = {4},
reduceSize = {4},
pool = nn.SpatialMaxPooling(3, 3, 1, 1, 1, 1),
transfer = nn.Tanh(),
})
save(net_inception_block, torch.rand(2, 3, 16, 16), 'net_inception_block', true)
local net_normalize = nn.Sequential()
net_normalize:add(nn.Normalize(2))
net_normalize:add(nn.Normalize(1, 1e-3))
net_normalize:add(nn.Normalize(2.7))
save(net_normalize, torch.rand(1, 24) * 3 - 0.5, 'net_normalize', true)
local net_padding = nn.Sequential()
net_padding:add(nn.Padding(2, 2, 4))
net_padding:add(nn.Padding(1, -1, 2))
net_padding:add(nn.Padding(3, -3, 3, 3))
net_padding:add(nn.SpatialZeroPadding(3, 1, 2, 0));
save(net_padding, torch.rand(2, 1, 3, 4), 'net_padding', true)
local net_spatial_zero_padding = nn.Sequential()
net_spatial_zero_padding:add(nn.SpatialZeroPadding(1, 0, 2, 3));
save(net_spatial_zero_padding, torch.rand(4, 2, 3), 'net_spatial_zero_padding', true)
-- OpenFace network.
-- require 'image'
-- torch.setdefaulttensortype('torch.FloatTensor')
-- net = torch.load('../openface_nn4.small2.v1.t7')
-- net:evaluate()
-- input = image.load('../../cv/shared/lena.png')
-- input = image.scale(input, 96, 96, 'simple'):reshape(1, 3, 96, 96)
-- output = net:forward(input):reshape(1, 128)
-- torch.save('net_openface_output.dat', output)
local net_spatial_reflection_padding = nn.Sequential()
net_spatial_reflection_padding:add(nn.SpatialReflectionPadding(5, 5, 5, 5));
save(net_spatial_reflection_padding, torch.rand(1, 3, 7, 8), 'net_spatial_reflection_padding', true)
local net_non_spatial = nn.Sequential()
net_non_spatial:add(nn.Dropout(0.3)) -- Dropout that replaces to an identity
net_non_spatial:add(nn.Linear(6, 5))
net_non_spatial:add(nn.Dropout(0.6, true)) -- Dropout that replaces to scale
net_non_spatial:add(nn.BatchNormalization(5)) -- 2D batch normalization
net_non_spatial:evaluate()
save(net_non_spatial, torch.rand(1, 6), 'net_non_spatial', true)
-- output = input + g ( f(input) * f(input) )
local net_residual = nn.Sequential()
net_residual:add(nn.ConcatTable()
:add(nn.Identity())
:add(nn.Sequential()
:add(nn.Sigmoid())
:add(nn.ConcatTable()
:add(nn.Identity())
:add(nn.Identity())
)
:add(nn.CAddTable())
:add(nn.Tanh())
)
)
:add(nn.CAddTable())
save(net_residual, torch.rand(2, 3, 4, 6), 'net_residual')
local net_spatial_zero_padding = nn.Sequential()
net_spatial_zero_padding:add(nn.SpatialUpSamplingNearest(2));
save(net_spatial_zero_padding, torch.rand(2, 3, 4, 5), 'net_spatial_upsampling_nearest')