If the purpose of dilated convolution is to extend receptive fields (extract image features from distant regions) and kernel 5x5 with mirror padding is also able to get the feature from distant regions. Why do people more often use the dilated convolution over kernel 5x5?
Thank you.
The architecture is based on the fact that dilated convolutions support exponential expansion of the receptive field without loss of resolution or coverage.
I think most people often use the dilated convolution is because it allows one to have larger receptive field with same computation and memory costs while also preserving resolution.
Also Dilated convolutions have generally improved performance. See in this paper Multi-Scale Context Aggregation by Dilated Convolutions we have better semantic segmentation results.
