coupe.composition-classifier/Composition Classifier/classification_test.lua at master · posgraph/coupe.composition-classifier · GitHub

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
require 'cunn'
require 'cudnn'
require 'nn'
require 'image'
require 'lfs'
require 'cutorch'
require 'optim'
-- cutorch.setDevice(1)
-- sofar = os.clock()

cmd = torch.CmdLine()
cmd:option('-net', '/media/eunbin/Data2/Dis_model/ggiruk/baseline-epoch-9.net', 'trained model')
cmd:option('-dir', '/home/eunbin/TorchNet/test/test_bad2', 'target directory')
opt = cmd:parse(arg or {})
num_bad = 0
num_good = 0
print(opt.net)
print(opt.dir)
model = torch.load(opt.net):cuda()
model:evaluate()

-- require 'hdf5'
-- print('Loading test data...')
-- testFile = hdf5.open('discriminator_test.h5', 'r')
-- testData = testFile:all()
-- testFile:close()

-- dataset = testData
-- data = dataset.data
-- label = dataset.label
-- dataset.size = data:size()[1]
-- batchSize = 32
-- classes = {'o', 'x'}
-- confusion = optim.ConfusionMatrix(classes)

-- print('<trainer> on testing Set:')
-- for t = 1,dataset.size,batchSize do
-- 	xlua.progress(t, dataset.size)
-- 	-- prepare input batch
-- 	local inputs = data:sub(t,math.min(t+batchSize-1, dataset.size))
-- 	inputs = inputs:cuda()
-- 	local targets = label:sub(t,math.min(t+batchSize-1, dataset.size))+1
-- 	--local targets_tl = targets:narrow(4,1,1)
-- 	--local targets_br = targets:narrow(4,2,1)
-- 	--local targets_both = {targets_tl, targets_br}
-- 	targets = targets:cuda()

-- 	local outputs = model:forward(inputs)
-- 	for i = 1,batchSize do
-- 		confusion:add(outputs[i]:view(-1), targets[i]:view(-1)[1])
-- 		print(outputs[i]:view(-1), targets[i]:view(-1)[1])
-- 		--confusion_br:add(outputs[2][i]:view(-1), targets_both[2][i]:view(-1)[1])
-- 	end
-- 	print(outputs)
-- end
-- print(confusion)
-- --print(confusion_br)
-- --testLogger:add{['% mean class accuracy (test set, top left)'] = confusion_tl.totalValid * 100,
-- --			   ['% mean class accuracy (test set, bottom right)'] = confusion_br.totalValid * 100}
-- confusion:zero()


for file in lfs.dir(opt.dir) do
if lfs.attributes(opt.dir..'/'..file, "mode") == "file" then
img = image.load(opt.dir..'/'..file, 3, 'float')
--transposed_img = img:transpose(2,3)
--imggg = image.vflip(transposed_img)
channel = img:size()[1]
height = img:size()[2]
width = img:size()[3]
output_size = channel * height * width

input = image.scale(img, '224x224', 'bicubic')
batch = input:view(1, table.unpack(input:size():totable()))
--input = input:view(1, 3, 224, 224)
--print(input)
--input = input:cuda()


output = model:forward(batch:cuda())
if output[1][1] > 0.5 then
	num_good = num_good + 1
	result = 1
else
	num_bad = num_bad + 1
	result = 0
end
--print(file..' : '.. result)


-- prob, idx = torch.max(output, 2)
-- if idx[1][1] == 1 then
-- 	num_good = num_good + 1
-- else
-- 	num_bad = num_bad + 1
-- end

 --print(file..' : '.. idx[1][1])

print(file..' : '.. result)

--print(output)
--print(idx)

end
end

print('number of good : '.. num_good)
print('number of bad : '.. num_bad)


--image.save('result.png', transposed_img)
--image.save('resultgg.png', imggg)

print('done.')