I would like to use Gaussian pyramid for pooling in convolution neural network. The target for this is to build a decovolution network to reconstruct the input(a image). That is to say when I obtain a convolution neural network model to generate some feature, then the original image can be reconstructed by the decovolution network. In the convnet, the common max pooling operation is non-invertible which is the reason to use Gaussian pyramid pooling. I have google the internet, but it seem that pyramid pooling is rarely used in convolution neural network.
So I would to know the difference between max pooling and pyramid pooling. What is the drawback of pyramid pooling in convolution neural network.
