In [13]:

    

require 'torch'
require 'nn'
require 'nnx'
require 'image' 
require 'optim'


    

In [2]:

    

-- download dataset
if not paths.dirp('cifar-10-batches-t7') then
   local www = 'http://torch7.s3-website-us-east-1.amazonaws.com/data/cifar-10-torch.tar.gz'
   local tar = paths.basename(www)
   os.execute('wget ' .. www .. '; '.. 'tar xvf ' .. tar)
end


    

In [3]:

    

trsize = 50000
tesize = 10000


    

In [4]:

    

-- load dataset
trainData = {
   data = torch.Tensor(50000, 3072),
   labels = torch.Tensor(50000),
   size = function() return trsize end
}
for i = 0,4 do
   subset = torch.load('cifar-10-batches-t7/data_batch_' .. (i+1) .. '.t7', 'ascii')
   trainData.data[{ {i*10000+1, (i+1)*10000} }] = subset.data:t()
   trainData.labels[{ {i*10000+1, (i+1)*10000} }] = subset.labels
end
trainData.labels = trainData.labels + 1

subset = torch.load('cifar-10-batches-t7/test_batch.t7', 'ascii')
testData = {
   data = subset.data:t():double(),
   labels = subset.labels[1]:double(),
   size = function() return tesize end
}
testData.labels = testData.labels + 1


    

In [5]:

    

-- resize dataset (if using small version)
trainData.data = trainData.data[{ {1,trsize} }]
trainData.labels = trainData.labels[{ {1,trsize} }]


    

In [6]:

    

testData.data = testData.data[{ {1,tesize} }]
testData.labels = testData.labels[{ {1,tesize} }]


    

In [7]:

    

-- reshape data
trainData.data = trainData.data:reshape(trsize,3,32,32)
testData.data = testData.data:reshape(tesize,3,32,32)


    

In [14]:

    

----------------------------------------------------------------------
-- preprocess/normalize train/test sets
--

print '<trainer> preprocessing data (color space + normalization)'

-- preprocess trainSet
normalization = nn.SpatialContrastiveNormalization(1, image.gaussian1D(7))
for i = 1,trainData:size() do
   -- rgb -> yuv
   local rgb = trainData.data[i]
   local yuv = image.rgb2yuv(rgb)
   -- normalize y locally:
   yuv[1] = normalization(yuv[{{1}}])
   trainData.data[i] = yuv
end
-- normalize u globally:
mean_u = trainData.data[{ {},2,{},{} }]:mean()
std_u = trainData.data[{ {},2,{},{} }]:std()
trainData.data[{ {},2,{},{} }]:add(-mean_u)
trainData.data[{ {},2,{},{} }]:div(-std_u)
-- normalize v globally:
mean_v = trainData.data[{ {},3,{},{} }]:mean()
std_v = trainData.data[{ {},3,{},{} }]:std()
trainData.data[{ {},3,{},{} }]:add(-mean_v)
trainData.data[{ {},3,{},{} }]:div(-std_v)

-- preprocess testSet
for i = 1,testData:size() do
   -- rgb -> yuv
   local rgb = testData.data[i]
   local yuv = image.rgb2yuv(rgb)
   -- normalize y locally:
   yuv[{1}] = normalization(yuv[{{1}}])
   testData.data[i] = yuv
end
-- normalize u globally:
testData.data[{ {},2,{},{} }]:add(-mean_u)
testData.data[{ {},2,{},{} }]:div(-std_u)
-- normalize v globally:
testData.data[{ {},3,{},{} }]:add(-mean_v)
testData.data[{ {},3,{},{} }]:div(-std_v)


    

    
Out[14]:





<trainer> preprocessing data (color space + normalization)	

In [15]:

    

print(testData.data:size())


    

In [16]:

    

print(trainData.data:size())


    

    
Out[16]:





 10000
     3
    32
    32
[torch.LongStorage of size 4]


 50000
     3
    32
    32
[torch.LongStorage of size 4]