danet

DANet

Dual Attention Network for Scene Segmentation

Introduction

Official Repo

Code Snippet

Abstract

In this paper, we address the scene segmentation task by capturing rich contextual dependencies based on the selfattention mechanism. Unlike previous works that capture contexts by multi-scale features fusion, we propose a Dual Attention Networks (DANet) to adaptively integrate local features with their global dependencies. Specifically, we append two types of attention modules on top of traditional dilated FCN, which model the semantic interdependencies in spatial and channel dimensions respectively. The position attention module selectively aggregates the features at each position by a weighted sum of the features at all positions. Similar features would be related to each other regardless of their distances. Meanwhile, the channel attention module selectively emphasizes interdependent channel maps by integrating associated features among all channel maps. We sum the outputs of the two attention modules to further improve feature representation which contributes to more precise segmentation results. We achieve new state-of-the-art segmentation performance on three challenging scene segmentation datasets, i.e., Cityscapes, PASCAL Context and COCO Stuff dataset. In particular, a Mean IoU score of 81.5% on Cityscapes test set is achieved without using coarse data. We make the code and trained model publicly available at this https URL.

Citation

@article{fu2018dual,
  title={Dual Attention Network for Scene Segmentation},
  author={Jun Fu, Jing Liu, Haijie Tian, Yong Li, Yongjun Bao, Zhiwei Fang,and Hanqing Lu},
  booktitle={The IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
  year={2019}
}

Results and models

Cityscapes

Method	Backbone	Crop Size	Lr schd	Mem (GB)	Inf time (fps)	mIoU	mIoU(ms+flip)	config	download
DANet	R-50-D8	512x1024	40000	7.4	2.66	78.74	-	config	model | log
DANet	R-101-D8	512x1024	40000	10.9	1.99	80.52	-	config	model | log
DANet	R-50-D8	769x769	40000	8.8	1.56	78.88	80.62	config	model | log
DANet	R-101-D8	769x769	40000	12.8	1.07	79.88	81.47	config	model | log
DANet	R-50-D8	512x1024	80000	-	-	79.34	-	config	model | log
DANet	R-101-D8	512x1024	80000	-	-	80.41	-	config	model | log
DANet	R-50-D8	769x769	80000	-	-	79.27	80.96	config	model | log
DANet	R-101-D8	769x769	80000	-	-	80.47	82.02	config	model | log

ADE20K

Method	Backbone	Crop Size	Lr schd	Mem (GB)	Inf time (fps)	mIoU	mIoU(ms+flip)	config	download
DANet	R-50-D8	512x512	80000	11.5	21.20	41.66	42.90	config	model | log
DANet	R-101-D8	512x512	80000	15	14.18	43.64	45.19	config	model | log
DANet	R-50-D8	512x512	160000	-	-	42.45	43.25	config	model | log
DANet	R-101-D8	512x512	160000	-	-	44.17	45.02	config	model | log

Pascal VOC 2012 + Aug

Method	Backbone	Crop Size	Lr schd	Mem (GB)	Inf time (fps)	mIoU	mIoU(ms+flip)	config	download
DANet	R-50-D8	512x512	20000	6.5	20.94	74.45	75.69	config	model | log
DANet	R-101-D8	512x512	20000	9.9	13.76	76.02	77.23	config	model | log
DANet	R-50-D8	512x512	40000	-	-	76.37	77.29	config	model | log
DANet	R-101-D8	512x512	40000	-	-	76.51	77.32	config	model | log