Guide To ResNeSt: A Better ResNet With The Same Costs

Convolution neural networks have largely dominated the computer vision domain, but in the last few years, feature-map attention architectures like SE-Net and SK-Net have started to assert dominance. In their paper “ResNeSt: Split-Attention Networks”,  Hang Zhang,  Chongruo Wu, et al. proposed a new ResNet variant that combines the best of both worlds. The ResNeSt architecture leverages the channel-wise attention with multi-path representation into a single unified Split-Attention block. It learns cross-channel feature correlations while preserving independent representation in the meta structure. Architecture & Approach Split-Attention block ResNeSt introduces the Split-Attention block, which enables feature-map attention across different feature-map groups. This Split-A