szagoruyko / imagenet-validation.torch Goto Github PK

Are you using "matio" anywhere in this repo?

I found the line

local matio = require 'matio'

in "main.lua", but couldn't find any other place in the repo where you'd be using it.

Hi Sergey,
I want to finetune the alexnet_torch.t7 model(https://gist.github.com/szagoruyko/dd032c529048492630fc) on a different dataset. Are the input channels to the net RGB, or BGR ? Also, which channels do the mean and std values stored in net.transform() correspond to ?
Thanks
{
mean :
{
1 : 0.48462227599918
2 : 0.45624044862054
3 : 0.40588363755159
}
std :
{
1 : 0.22889466674951
2 : 0.22446679341259
3 : 0.22495548344775
}
}

I was thinking that, for completeness, you may want to train a GoogLeNet as well.
Here's the model.

Hi, Is it possible to make the torch alexnet model public ? I notice that , that model obtains a top-1 accuracy of 56.5% which is closer to the actual alexnet model compared to the ported bvlc_alexnet model which obtains ~54%. This would be very helpful. Thanks!