Error rate reached : logprob: 0.066761, 0.002564 is it ok or smth went wrong ?

[teknosains]

Haloo

I used your dropout version of Convnet for Food classification.

I have 8 batches :

1-6 : train, 7 validation
and 8 for test

I followed the methodology https://code.google.com/p/cuda-convnet/wiki/Methodology

I started the training :
--train-range=1-6 --test-range=7 --epochs=200

an so on like the methodology...

The error rates for each batches is Good ( around 0.11)

but when I test it ( last method) with batch 8 . its 63% (final result).

I dont know what caused this. I assume that Foods share almost the same colors, shape, etc, mayble those are the cause.

What do you suggest ?

Thanks