What is dilated convolution and why is it used?
Updated May 15, 2026
Short answer
Dilated convolution expands receptive field without increasing parameters.
Deep explanation
It introduces gaps between kernel elements, allowing the network to capture broader context while preserving resolution. Used in segmentation tasks.
Real-world example
Used in DeepLab models for semantic segmentation.
Common mistakes
- Confusing dilation with pooling-based downsampling.
Follow-up questions
- What is receptive field?
- Why avoid downsampling?