What is dilated convolution and how does it help in CNN architectures?

Dilated convolution expands the receptive field without increasing the number of parameters by inserting gaps between kernel elements.

Dilated (atrous) convolution introduces spaces between kernel elements, allowing the filter to cover a larger area without increasing kernel size or computation. This helps capture multi-scale context, especially useful in segmentation tasks. It preserves resolution while increasing receptive field exponentially with depth.