Stride for pooling layer

[always...@gmail.com]

Dear Sir

I have question regarding the pooling layer.

If I set 

net.layers{end+1}.properties = struct('type',2,'numFM',10,'kernel', [5 5 5]  'stride', [2 2 2], 'pooling, [2 2 2]), 

does this mean i use 10 kernels with size of [5 5 5] with stride [5 5 5] , then I perform max pooling with size of [2 2 2] and with stride of [ 5 5 5]??

does stride [ 5 5 5] work for both convolution as well as maxpooling? or there are different stride for max pooling?

I could not find the answer in the internet so i am asking this question.

Thank you again

Best regard

Hang-Rai Kim

[Hagay Garty]

maxpooling works on the output of the network layer. 

The order is 1) convolution, 2) stride, 3) max pooling.

In your example the input is convolved with the 5x5x5 kernels, using stride 2x2x2. Then max-pooling of size 2x2x2 is applied.

stride and max-pooling are configured as a property of a convolutional layer and not as a separate layer.