This should probably have similar options as a normal tensorflow convolutional layer. (padding, stride, etc)