Convolution operation can be replaced by MaxPool

[miguelvr]

Description:

A maxpool operation can be used for the block mask calculation and might be more efficient to compute than a convolution.

[linzehui]

have you figured out how to replace convolution with maxpooling operation? I found convolution operation very slow.

[linzehui]

I tested two operation, MaxPooling can be much faster than Convolution.
An example may be like that:
mask = F.max_pool2d(input=mask[:, None, :, :],
kernel_size= (block_size, block_size),
stride=(1,1),
padding=(block_size ,block_size))

[miguelvr]

I know how to do it, just don't have much time to do it. Can you do a PR?