gcnet

GCNet: Non-local Networks Meet Squeeze-Excitation Networks and Beyond

Introduction

@inproceedings{cao2019gcnet,
  title={Gcnet: Non-local networks meet squeeze-excitation networks and beyond},
  author={Cao, Yue and Xu, Jiarui and Lin, Stephen and Wei, Fangyun and Hu, Han},
  booktitle={Proceedings of the IEEE International Conference on Computer Vision Workshops},
  pages={0--0},
  year={2019}
}

Results and models

Cityscapes

Method	Backbone	Crop Size	Lr schd	Mem (GB)	Inf time (fps)	mIoU	mIoU(ms+flip)	download
GCNet	R-50-D8	512x1024	40000	5.8	3.93	77.69	78.56	model | log
GCNet	R-101-D8	512x1024	40000	9.2	2.61	78.28	79.34	model | log
GCNet	R-50-D8	769x769	40000	6.5	1.67	78.12	80.09	model | log
GCNet	R-101-D8	769x769	40000	10.5	1.13	78.95	80.71	model | log
GCNet	R-50-D8	512x1024	80000	-	-	78.48	80.01	model | log
GCNet	R-101-D8	512x1024	80000	-	-	79.03	79.84	model | log
GCNet	R-50-D8	769x769	80000	-	-	78.68	80.66	model | log
GCNet	R-101-D8	769x769	80000	-	-	79.18	80.71	model | log

ADE20K

Method	Backbone	Crop Size	Lr schd	Mem (GB)	Inf time (fps)	mIoU	mIoU(ms+flip)	download
GCNet	R-50-D8	512x512	80000	8.5	23.38	41.47	42.85	model | log
GCNet	R-101-D8	512x512	80000	12	15.20	42.82	44.54	model | log
GCNet	R-50-D8	512x512	160000	-	-	42.37	43.52	model | log
GCNet	R-101-D8	512x512	160000	-	-	43.69	45.21	model | log

Pascal VOC 2012 + Aug

Method	Backbone	Crop Size	Lr schd	Mem (GB)	Inf time (fps)	mIoU	mIoU(ms+flip)	download
GCNet	R-50-D8	512x512	20000	5.8	23.35	76.42	77.51	model | log
GCNet	R-101-D8	512x512	20000	9.2	14.80	77.41	78.56	model | log
GCNet	R-50-D8	512x512	40000	-	-	76.24	77.63	model | log
GCNet	R-101-D8	512x512	40000	-	-	77.84	78.59	model | log