Pooling is a downsampling operation that reduces the spatial dimensions of feature maps while retaining important information. It operates on non-overlapping regions.
Key Characteristics:
• Parameter-free: No learnable weights
• Translation invariant: Small shifts don't change output much
• Dimensionality reduction: Reduces spatial size
• Feature selection: Keeps most important activations
✓
Reduces computational complexity
✓
Provides translation invariance
✓
Controls overfitting
✓
Increases receptive field
🎯 Types of Pooling
🔥 Max Pooling
Takes the maximum value from each region. Preserves the strongest activations and provides translation invariance.
📊 Average Pooling
Computes the average of all values in each region. Provides smoother downsampling with less information loss.
🌍 Global Pooling
Pools over the entire feature map, reducing it to a single value per channel. Used before final classification layers.
🎛️ Adaptive Pooling
Automatically adjusts pooling to produce a fixed output size regardless of input dimensions.
Interactive Pooling Demonstration
Input Feature Map (4×4)
→
Max Pooling (2×2, stride=2)
Click "Play" or "Next" to see pooling step by step