saksenadhruv/cuda-convnet

Getting nan values while using response normalization layer

Opened this issue · 1 comments

The network training is fine without adding any contrast normalization layer 
(all types), but ones add the contrast normalization layers, after several 
iterations the net gets nan values.  I tried different values of the size, 
scale and pow values, and tried to place the layer before and after pooling 
layer. 


Original issue reported on code.google.com by zybzm...@gmail.com on 8 May 2013 at 7:44

I have a similar issue. Does it still work when you remove the contrast 
normalization layer? 

Original comment by olivern...@gmail.com on 13 Sep 2013 at 7:30