juniorCNN
What is stride in CNN and how does it affect feature extraction?
Updated May 15, 2026
Short answer
Stride defines how many pixels the filter moves at each step, controlling output size and computation cost.
Deep explanation
Stride determines the movement of the convolution filter across the input image. A stride of 1 moves pixel by pixel, producing detailed feature maps. Higher strides downsample the image faster but may lose fine-grained information. It directly affects spatial resolution and computational efficiency.
Real-world example
Used in early layers of CNNs to reduce image size in mobile vision models.
Common mistakes
- Using large stride values too early and losing important spatial details.
Follow-up questions
- How does stride differ from pooling?
- What happens when stride equals kernel size?