ResNeSt

lastmod 2026-11-10created 2021-01-202 edits

ResNeSt is a convolutional neural network for image classification that adds Split-Attention blocks to the ResNet architecture. It splits feature maps into multiple groups (cardinal groups) and splits, then applies attention weights to each split and combines them, increasing representational power without greatly increasing computation. Besides ImageNet classification, it has also been used as a backbone for object detection and segmentation models.

ResNeSt: Split-Attention Networks

See also

• Deep learning

Documentation

ResNeSt - Split-Attention Networks

https://arxiv.org/abs/2004.08955v1

Favorite site

• ResNeSt: Split-Attention Networks | Papers With Code
• Github - ResNeSt