zhanghang1989/ResNeSt

Split-attention modules partition channel groups and apply learned attention weights to aggregate multi-scale representations within ResNet bottleneck blocks. Available as PyTorch and MXNet/Gluon implementations with pretrained weights, it integrates directly with Detectron2, MMDetection, and semantic segmentation frameworks (PyTorch Encoding, GluonCV) for transfer learning on downstream tasks like object detection and panoptic segmentation.

3,264 stars and 11,896 monthly downloads. No commits in the last 6 months. Available on PyPI.