-
Notifications
You must be signed in to change notification settings - Fork 0
/
apascal_svm_train.lua
313 lines (281 loc) · 10.3 KB
/
apascal_svm_train.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
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
require 'data_read'
require 'loadcaffe'
require 'cutorch'
require 'cunn'
require 'optim'
local matio = require 'matio'
fname = sys.fpath()
cmd = torch.CmdLine()
cmd:option('-save', fname:gsub('.lua',''), 'subdirectory to save/log experiments in')
opt1 = cmd:parse(arg)
opt = lapp[[
--batchSize (default 64) Sub-batch size
--dataRoot (default ./home/yang/abd/overfeat-torch/aPascal-aYahoo/) Data root folder
--imageRoot (default ./home/yang/abd/overfeat-torch/aPascal-aYahoo/apascal_images/) Image dir
--testRoot (default ./home/yang/abd/overfeat-torch/aPascal-aYahoo/ayahoo_test_images/) test dir
]]
torch.setdefaulttensortype('torch.FloatTensor')
torch.manualSeed(1)
dataset = Dataset('/home/yang/abd/overfeat-torch/aPascal-aYahoo/',
'/home/yang/abd/overfeat-torch/aPascal-aYahoo/apascal_images/',
'/home/yang/abd/overfeat-torch/aPascal-aYahoo/ayahoo_test_images/',
torch.Tensor({0.0,0,0,0,0}),1.0,
opt.batchSize
)
train_size, val_size = dataset:size()
local model, sgdState
--------------------------------------------------------Building Model
model = nn.Sequential()
model = torch.load('/home/yang/abd/overfeat-torch/model.t7')
model:cuda()
print(model)
--------------------------------------------------------Parameter initialize
sgdState = {
--- For SGD with momentum ---
learningRate = 1e-3,
learningRateDecay = 1e-7,
weightDecay = 1e-4,
momentum = 0.9,
}
mean = torch.Tensor(1,3,231,231):float()
channels = {'r','g','b'}
mean = {0.411,0.3812,0.3574}
std = {0.2688,0.2582,0.2612}
---------------------------------------
-- Feature extraction : Training images
---------------------------------------
local co = dataset:get_train_sample_co()
train_num, valid_num = dataset:size()
features = torch.Tensor(train_num,4096):float()
train_labels = {}
tr_label = torch.Tensor(train_num,64):float()
for label_i = 1,64 do
train_labels[label_i] = torch.Tensor(train_num):zero():float()
end
index = 1
while true do
flag,ims,target_labels = coroutine.resume(co,dataset)
if( ims == nil ) then
break
end
for i, channel in ipairs(channels) do
ims[{ {},i,{},{} }]:add(-mean[i])
ims[{ {},i,{},{} }]:div(std[i])
end
processed = ims
collectgarbage();collectgarbage();
ret = model:forward(processed:cuda())
features:sub(index,index + ims:size()[1]-1):copy(model['modules'][16].output:float()) --FC7 features
for label_i = 1,64 do
train_labels[label_i]:sub(index,index + ims:size()[1]-1):copy(target_labels[label_i])
end
index = index + ims:size()[1]
xlua.progress(index,train_num)
end
for i = 1,64 do
tr_label[{{},i}]:copy(train_labels[i])
end
matio.save('train-fc7-features_mn.mat',{train_features = features})
matio.save('train-attribute-label_mn.mat',tr_label)
torch.save( '/home/yang/abd/overfeat-torch/experiments_svm/train-fc7-features_mn.th', features )
torch.save('/home/yang/abd/overfeat-torch/experiments_svm/train-attribute-labels_mn.th', train_labels )
-----------------------------------------
-- Save to appropriate format for SVM : Training Images
------------------------------------------
train_num, valid_num = dataset:size()
features = torch.load( '/home/yang/abd/overfeat-torch/experiments_svm/train-fc7-features_mn.th')
labels = torch.load( '/home/yang/abd/overfeat-torch/experiments_svm/train-attribute-labels_mn.th')
--
files = {}
for i = 1,64 do
file = io.open( '/home/yang/abd/overfeat-torch/experiments_svm/attr'..tostring(i)..'.txt','w+')
files[i] = file
end
for i = 1, train_num do
local t = {}
for j = 1,4096 do
table.insert(t,tostring(j)..':'..tostring(features[i][j])..' ')
end
s = table.concat(t,'')..'\n'
for label_i = 1, 64 do
if( labels[label_i][i] == 0 ) then
files[label_i]:write('-1'..' '..s)
else
files[label_i]:write('+1'..' '..s)
end
end
xlua.progress(i,train_num)
end
for i = 1,64 do
files[i]:close()
end
---------------------------------------
-- Feature extraction : Validation set images
---------------------------------------
--local co = dataset:get_valid_sample_co()
--train_num, valid_num = dataset:size()
--features = torch.Tensor(valid_num,4096):float()
--labels = {}
--for label_i = 1,64 do
-- labels[label_i] = torch.Tensor(valid_num):zero():float()
--end
--
--index = 1
--while true do
-- flag,ims,target_labels = coroutine.resume(co,dataset)
-- if( ims == nil ) then
-- break
-- end
-- processed = ims*255 - torch.expand(mean,ims:size()[1],3,231,231)
--
-- collectgarbage();collectgarbage();
-- ret = model:forward(processed:cuda())
-- features:sub(index,index + ims:size()[1]-1):copy(model['modules'][16].output:float()) --FC7 features
-- for label_i = 1,64 do
-- labels[label_i]:sub(index,index + ims:size()[1]-1):copy(target_labels[label_i])
-- end
-- index = index + ims:size()[1]
--
-- xlua.progress(index,valid_num)
--end
--torch.save('/home/dhill/abd/overfeat-torch/experiments_svm/valid-fc7-features.th', features )
--torch.save('/home/dhill/abd/overfeat-torch/experiments_svm/valid-attribute-labels.th', labels )
-----------------------------------------
-- Save to appropriate format for SVM : Validation set Images
-----------------------------------------
--train_num, valid_num = dataset:size()
--features = torch.load('/home/dhill/abd/overfeat-torch/experiments_svm/valid-fc7-features.th')
--labels = torch.load('/home/dhill/abd/overfeat-torch/experiments_svm/valid-attribute-labels.th')
--files = {}
--for i = 1,64 do
-- file = io.open('/home/dhill/abd/overfeat-torch/experiments_svm/valid-attr'..tostring(i)..'.txt','w+')
-- files[i] = file
--end
--for i = 1, valid_num do
-- local t = {}
-- for j = 1,4096 do
-- table.insert(t,tostring(j)..':'..tostring(features[i][j])..' ')
-- end
-- s = table.concat(t,'')..'\n'
-- for label_i = 1, 64 do
-- if( labels[label_i][i] == 0 ) then
-- files[label_i]:write('-1'..' '..s)
-- else
-- files[label_i]:write('+1'..' '..s)
-- end
-- end
-- xlua.progress(i,valid_num)
--end
--for i = 1,64 do
-- files[i]:close()
--end
--------------------------------------
-- Feature extraction : Testing set images
---------------------------------------
local co = dataset:get_test_sample_co()
train_num,valid_num,test_num = dataset:size()
print(test_num)
features = torch.Tensor(test_num,4096):float()
test_labels = {}
test_labels1 = torch.Tensor(test_num,64):float()
for label_i = 1,64 do
test_labels[label_i] = torch.Tensor(test_num):zero():float()
end
index = 1
while true do
flag,ims,target_labels = coroutine.resume(co,dataset)
if( ims == nil ) then
break
end
for i, channel in ipairs(channels) do
ims[{ {},i,{},{} }]:add(-mean[i])
ims[{ {},i,{},{} }]:div(std[i])
end
processed = ims
collectgarbage();collectgarbage();
ret = model:forward(processed:cuda())
features:sub(index,index + ims:size()[1]-1):copy(model['modules'][16].output:float()) --FC7 features
for label_i = 1,64 do
test_labels[label_i]:sub(index,index + ims:size()[1]-1):copy(target_labels[label_i])
end
index = index + ims:size()[1]
xlua.progress(index,test_num)
end
for i = 1,64 do
test_labels1[{{},i}]:copy(test_labels[i])
end
matio.save('test-fc7-features_mn.mat',{test_features = features})
matio.save('test-attribute-label_mn.mat',test_labels1)
torch.save('/home/yang/abd/overfeat-torch/experiments_svm/test-fc7-features_mn.th', features )
torch.save('/home/yang/abd/overfeat-torch/experiments_svm/test-attribute-labels_mn.th', test_labels )
-----------------------------------------
-- Save to appropriate format for SVM : Testing set Images
-----------------------------------------
train_num, valid_num,test_num = dataset:size()
features = torch.load('/home/yang/abd/overfeat-torch/experiments_svm/test-fc7-features_mn.th')
labels = torch.load('/home/yang/abd/overfeat-torch/experiments_svm/test-attribute-labels_mn.th')
files = {}
for i = 1,64 do
file = io.open('/home/yang/abd/overfeat-torch/experiments_svm/test-attr'..tostring(i)..'.txt','w+')
files[i] = file
end
for i = 1, test_num do
local t = {}
for j = 1,4096 do
table.insert(t,tostring(j)..':'..tostring(features[i][j])..' ')
end
s = table.concat(t,'')..'\n'
for label_i = 1, 64 do
if( labels[label_i][i] == 0 ) then
files[label_i]:write('-1'..' '..s)
else
files[label_i]:write('+1'..' '..s)
end
end
xlua.progress(i,test_num)
end
for i = 1,64 do
files[i]:close()
end
---------------------------------------------------
-- Run SVM
---------------------------------------------------
true_positive = torch.Tensor(64):zero()
true_negative = torch.Tensor(64):zero()
false_positive = torch.Tensor(64):zero()
false_negative = torch.Tensor(64):zero()
accuracy_valid = torch.Tensor(64):zero()
accuracy_test = torch.Tensor(64):zero()
prediction = torch.Tensor(64,2642):zero()
svm = require 'svm'
for label_i = 1,64 do
collectgarbage(); collectgarbage();
train_data = svm.ascread('/home/dhill/abd/overfeat-torch/experiments_svm/attr'..tostring(label_i)..'.txt')
-- valid_data = svm.ascread('/home/dhill/abd/overfeat-torch/experiments_svm/valid-attr'..tostring(label_i)..'.txt')
test_data = svm.ascread('/home/dhill/abd/overfeat-torch/experiments_svm/test-attr'..tostring(label_i)..'.txt')
model = liblinear.train(train_data)
labels_t,accuracy_t,dec_t = liblinear.predict(test_data,model)--['-b 1'])
print(model)
accuracy_test[{label_i}] = accuracy_t[1]
for i = 1, labels_t:size()[1] do
if( labels_t[i] > 0 ) then
if( test_data[i][1] > 0 ) then
true_positive[label_i] = true_positive[label_i] + 1
else
false_positive[label_i] = false_positive[label_i] + 1
end
else
if( test_data[i][1] > 0 ) then
false_negative[label_i] = false_negative[label_i] + 1
else
true_negative[label_i] = true_negative[label_i] + 1
end
end
end
print'checking...'
prediction[{label_i,{}}]:copy(dec_t)
xlua.progress(label_i,64)
end
acc_t = (accuracy_test:sum())/64
print(acc_t)]]
matio.save('Data_auc_svm_dec.mat',{scores_svm = prediction})