resnest

ResNeSt

ResNeSt: Split-Attention Networks

Abstract

It is well known that featuremap attention and multi-path representation are important for visual recognition. In this paper, we present a modularized architecture, which applies the channel-wise attention on different network branches to leverage their success in capturing cross-feature interactions and learning diverse representations. Our design results in a simple and unified computation block, which can be parameterized using only a few variables. Our model, named ResNeSt, outperforms EfficientNet in accuracy and latency trade-off on image classification. In addition, ResNeSt has achieved superior transfer learning results on several public benchmarks serving as the backbone, and has been adopted by the winning entries of COCO-LVIS challenge.

Results and Models

Faster R-CNN

Backbone	Style	Lr schd	Mem (GB)	Inf time (fps)	box AP	Config	Download
S-50-FPN	pytorch	1x	4.8	-	42.0	config	model | log
S-101-FPN	pytorch	1x	7.1	-	44.5	config	model | log

Mask R-CNN

Backbone	Style	Lr schd	Mem (GB)	Inf time (fps)	box AP	mask AP	Config	Download
S-50-FPN	pytorch	1x	5.5	-	42.6	38.1	config	model | log
S-101-FPN	pytorch	1x	7.8	-	45.2	40.2	config	model | log

Cascade R-CNN

Backbone	Style	Lr schd	Mem (GB)	Inf time (fps)	box AP	Config	Download
S-50-FPN	pytorch	1x	-	-	44.5	config	model | log
S-101-FPN	pytorch	1x	8.4	-	46.8	config	model | log

Cascade Mask R-CNN

Backbone	Style	Lr schd	Mem (GB)	Inf time (fps)	box AP	mask AP	Config	Download
S-50-FPN	pytorch	1x	-	-	45.4	39.5	config	model | log
S-101-FPN	pytorch	1x	10.5	-	47.7	41.4	config	model | log

Citation

@article{zhang2020resnest,
title={ResNeSt: Split-Attention Networks},
author={Zhang, Hang and Wu, Chongruo and Zhang, Zhongyue and Zhu, Yi and Zhang, Zhi and Lin, Haibin and Sun, Yue and He, Tong and Muller, Jonas and Manmatha, R. and Li, Mu and Smola, Alexander},
journal={arXiv preprint arXiv:2004.08955},
year={2020}
}