I believe we are using WRONG mean-std prepossessing to CIFAR-100

[dlmacedo]

Dear Friends,

I believe we are using the wrong mean std preprocessing to CIFAR-100 since the parameters in the code are for CIFAR-10 dataset.

David

[dlmacedo]

--##
--[[
local meanstd = {mean = {125.3, 123.0, 113.9}, std  = {63.0,  62.1,  66.7}}
--]]
if opt.dataset == "./datasets/cifar10_original.t7" then
  meanstd = {mean = {125.3, 123.0, 113.9}, std  = {63.0,  62.1,  66.7}, dataset = "cifar-10"}
elseif opt.dataset == "./datasets/cifar100_original.t7" then
  meanstd = {mean = {129.3, 124.1, 112.4}, std  = {68.2,  65.4,  70.4}, dataset = "cifar-100"}
end
--##

print(c.blue '==>' ..' loading data')
local provider = torch.load(opt.dataset)
opt.num_classes = provider.testData.labels:max()
if torch.type(provider.trainData.data) == 'torch.ByteTensor' then
   --##
   print("##########")
   print(meanstd.dataset)
   print("##########")
   --##
   for i,v in ipairs{'trainData', 'testData'} do
      provider[v].data = provider[v].data:float()--:div(256)
      for ch=1,3 do
         provider[v].data:select(2,ch):add(-meanstd.mean[ch]):div(meanstd.std[ch])
      end
   end
end