how to use stride in customized pooling layer?

I am trying to write a Customized pooling layer. I am stuck at stride.

Here is my code

classdef meanPoolingLayer < nnet.layer.Layer
  
    properties (Learnable)
        
    end
    methods
        function layer = meanPoolingLayer(name,stride,padding,poolsize,paddingmode)
            % (Optional) Create a myLayer.
            % This function must have the same name as the class.
            layer.Name = name;
            layer.Stride=stride;
            layer.Padding=padding;
            layer.PoolSize=poolsize;
            layer.PaddingMode=paddingmode;
            % Layer constructor function goes here.
        end
        
        function [Z1] = predict(layer, X1)
        
           X=double(X1);
%        Z1=mean(mean(X1));
      end
   end
end

How to slide window over entire data for this pooling.

Thanks in advance

Answer 1

[Ayush Aniket]

Hi Nayab,

To implement the striding functionality, you need to manually iterate over the input feature map and apply the mean pooling operation to each window defined by your stride and pool size. This involves a bit of manual calculation to determine how to move the window across the input data, taking into account the stride, pool size, and optionally, padding.

Here's an example of mean pooling layer with stride and padding (assuming 'valid' padding for simplicity):

function Z = predict(layer, X)
            % Predict function for performing mean pooling.
            
            [h, w, c, n] = size(X); % Assuming 'X' is a 4-D array: height, width, channels, batch size
            
            % Calculate output dimensions
            outHeight = floor((h - layer.PoolSize(1)) / layer.Stride(1)) + 1;
            outWidth = floor((w - layer.PoolSize(2)) / layer.Stride(2)) + 1;
            
            % Initialize output
            Z = zeros(outHeight, outWidth, c, n, 'like', X);
            
            % Apply mean pooling
            for i = 1:outHeight
                for j = 1:outWidth
                    for k = 1:c
                        for l = 1:n
                            % Define window boundaries
                            rowStart = (i-1)*layer.Stride(1) + 1;
                            rowEnd = rowStart + layer.PoolSize(1) - 1;
                            colStart = (j-1)*layer.Stride(2) + 1;
                            colEnd = colStart + layer.PoolSize(2) - 1;
                            
                            % Extract window and compute mean
                            window = X(rowStart:rowEnd, colStart:colEnd, k, l);
                            Z(i, j, k, l) = mean(window, 'all');
                        end
                    end
                end
            end

The method calculates the output size based on the input dimensions, pool size, and stride, then iterates over the input to apply mean pooling to each window.