Large Stride : To reduce the spatial extent of the feature maps in the ConvNets, we can use the larger stride. Although this reduces the spatial extent, it is a very aggressive, in that we loose the information.

Pooling :  With pooling, we take small strides and then somehow for all the convolution in the neighborhood, combine somehow . This is called Pooling.

large stride vs Pooling Convolution Networks.JPG


max pool