soit

SOIT: Segmenting Objects with Instance-Aware Transformers

Introduction

A fully end-to-end instance segmentation method based on Transformers.

Results and Models

Backbone	Lr schd	Dataset	Test scale	mAP	AP50	AP75	APS	APM	APL	mAPbox	Config	Download
R-50	50e	COCO	(1333, 800)	42.2	64.6	45.3	23.1	45.3	61.8	48.9	config	Google Drive | BaiduYun
R-101	50e	COCO	(1333, 800)	42.9	65.7	46.0	23.1	46.4	63.3	49.5	config	Google Drive | BaiduYun

NOTE

1. AP without superscript denotes mask AP. mAP^box denotes bbox AP.

Citation

@inproceedings{yu2022soit,
  title={SOIT: Segmenting Objects with Instance-Aware Transformers},
  author={Yu, Xiaodong and Shi, Dahu and Wei, Xing and Ren, Ye and Ye, Tingqun and Tan, Wenming},
  booktitle={Proceedings of the AAAI Conference on Artificial Intelligence},
  year={2022}
}