wing-ops/prediction/image/mx15hdi/Detect/mmsegmentation/configs/nonlocal_net

NonLocal Net

Non-local Neural Networks

Introduction

Official Repo

Code Snippet

Abstract

Both convolutional and recurrent operations are building blocks that process one local neighborhood at a time. In this paper, we present non-local operations as a generic family of building blocks for capturing long-range dependencies. Inspired by the classical non-local means method in computer vision, our non-local operation computes the response at a position as a weighted sum of the features at all positions. This building block can be plugged into many computer vision architectures. On the task of video classification, even without any bells and whistles, our non-local models can compete or outperform current competition winners on both Kinetics and Charades datasets. In static image recognition, our non-local models improve object detection/segmentation and pose estimation on the COCO suite of tasks. Code is available at this https URL.

Citation

@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)	config	download
NonLocalNet	R-50-D8	512x1024	40000	7.4	2.72	78.24	-	config	model | log
NonLocalNet	R-101-D8	512x1024	40000	10.9	1.95	78.66	-	config	model | log
NonLocalNet	R-50-D8	769x769	40000	8.9	1.52	78.33	79.92	config	model | log
NonLocalNet	R-101-D8	769x769	40000	12.8	1.05	78.57	80.29	config	model | log
NonLocalNet	R-50-D8	512x1024	80000	-	-	78.01	-	config	model | log
NonLocalNet	R-101-D8	512x1024	80000	-	-	78.93	-	config	model | log
NonLocalNet	R-50-D8	769x769	80000	-	-	79.05	80.68	config	model | log
NonLocalNet	R-101-D8	769x769	80000	-	-	79.40	80.85	config	model | log

ADE20K

Method	Backbone	Crop Size	Lr schd	Mem (GB)	Inf time (fps)	mIoU	mIoU(ms+flip)	config	download
NonLocalNet	R-50-D8	512x512	80000	9.1	21.37	40.75	42.05	config	model | log
NonLocalNet	R-101-D8	512x512	80000	12.6	13.97	42.90	44.27	config	model | log
NonLocalNet	R-50-D8	512x512	160000	-	-	42.03	43.04	config	model | log
NonLocalNet	R-101-D8	512x512	160000	-	-	44.63	45.79	config	model | log

Pascal VOC 2012 + Aug

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