How Does Pooling Transform Images? A Visual Guide to Max, Min & Average Pooling

Pooling, or "池化" in Chinese, is a common operation in neural networks, especially convolutional neural networks (CNNs). The most frequently used variant is Max Pooling, which selects the maximum value within a sliding window.

Pooling windows are not limited to 2×2; sizes such as 3×3, 5×5, etc., are also valid, and the stride (step size) can be adjusted accordingly. Besides neural networks, pooling can be used to create a mosaic effect on images.

Applying Max Pooling to an example image yields the following result:

Min Pooling and Average Pooling work similarly, differing only in the aggregation function used.

Min Pooling example:

Average Pooling example (which often looks more natural, though results may vary with different images):

Below are Python implementations of the three pooling operations. Since the images have multiple color channels, remember to process each channel separately.

import cv2
import numpy as np

def average_pooling(img, G=5):
    out = img.copy()
    H, W, C = img.shape
    Nh = H // G
    Nw = W // G
    for y in range(Nh):
        for x in range(Nw):
            for c in range(C):
                out[G*y:G*(y+1), G*x:G*(x+1), c] = np.mean(out[G*y:G*(y+1), G*x:G*(x+1), c]).astype(np.int64)
    return out

def max_pooling(img, G=5):
    out = img.copy()
    H, W, C = img.shape
    Nh = H // G
    Nw = W // G
    for y in range(Nh):
        for x in range(Nw):
            for c in range(C):
                out[G*y:G*(y+1), G*x:G*(x+1), c] = np.max(out[G*y:G*(y+1), G*x:G*(x+1), c]).astype(np.int64)
    return out

def min_pooling(img, G=5):
    out = img.copy()
    H, W, C = img.shape
    Nh = H // G
    Nw = W // G
    for y in range(Nh):
        for x in range(Nw):
            for c in range(C):
                out[G*y:G*(y+1), G*x:G*(x+1), c] = np.min(out[G*y:G*(y+1), G*x:G*(x+1), c]).astype(np.int64)
    return out