nonlocal_net

Non-local Neural Networks

Introduction

@inproceedings{wang2018non,
  title={Non-local neural networks},
  author={Wang, Xiaolong and Girshick, Ross and Gupta, Abhinav and He, Kaiming},
  booktitle={Proceedings of the IEEE conference on computer vision and pattern recognition},
  pages={7794--7803},
  year={2018}
}

Results and models

Cityscapes

Method	Backbone	Crop Size	Lr schd	Mem (GB)	Inf time (fps)	mIoU	mIoU(ms+flip)	download
NonLocal	R-50-D8	512x1024	40000	7.4	2.72	78.24	-	model | log
NonLocal	R-101-D8	512x1024	40000	10.9	1.95	78.66	-	model | log
NonLocal	R-50-D8	769x769	40000	8.9	1.52	78.33	79.92	model | log
NonLocal	R-101-D8	769x769	40000	12.8	1.05	78.57	80.29	model | log
NonLocal	R-50-D8	512x1024	80000	-	-	78.01	-	model | log
NonLocal	R-101-D8	512x1024	80000	-	-	78.93	-	model | log
NonLocal	R-50-D8	769x769	80000	-	-	79.05	80.68	model | log
NonLocal	R-101-D8	769x769	80000	-	-	79.40	80.85	model | log

ADE20K

Method	Backbone	Crop Size	Lr schd	Mem (GB)	Inf time (fps)	mIoU	mIoU(ms+flip)	download
NonLocal	R-50-D8	512x512	80000	9.1	21.37	40.75	42.05	model | log
NonLocal	R-101-D8	512x512	80000	12.6	13.97	42.90	44.27	model | log
NonLocal	R-50-D8	512x512	160000	-	-	42.03	43.04	model | log
NonLocal	R-101-D8	512x512	160000	-	-	43.36	44.83	model | log

Pascal VOC 2012 + Aug

Method	Backbone	Crop Size	Lr schd	Mem (GB)	Inf time (fps)	mIoU	mIoU(ms+flip)	download
NonLocal	R-50-D8	512x512	20000	6.4	21.21	76.20	77.12	model | log
NonLocal	R-101-D8	512x512	20000	9.8	14.01	78.15	78.86	model | log
NonLocal	R-50-D8	512x512	40000	-	-	76.65	77.47	model | log
NonLocal	R-101-D8	512x512	40000	-	-	78.27	79.12	model | log