Calculate the output feature map size after an average pooling operation.
Average pooling computes the mean value within each pooling window, providing smoother downsampling than max pooling. The output size formula is identical to max pooling: floor((input + 2×padding − pool_size) / stride) + 1.
Avg Pool Output
output = floor((input + 2*padding - pool_size) / stride) + 1
Average pooling is preferred for global aggregation (Global Average Pooling before classifiers), in tasks where all feature values contribute (not just peaks), and in some GAN architectures.
GAP uses pool_size equal to the spatial dimension, collapsing each channel to a single number. It replaces fully connected layers at the end of CNNs, reducing overfitting.