Skip to content

Latest commit

 

History

History
67 lines (48 loc) · 14.4 KB

README.md

File metadata and controls

67 lines (48 loc) · 14.4 KB
Raw

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.

Results and models

Cityscapes

Method Backbone Crop Size Lr schd Mem (GB) Inf time (fps) Device mIoU mIoU(ms+flip) config download
DANet R-50-D8 512x1024 40000 7.4 2.66 V100 78.74 - config model | log
DANet R-101-D8 512x1024 40000 10.9 1.99 V100 80.52 - config model | log
DANet R-50-D8 769x769 40000 8.8 1.56 V100 78.88 80.62 config model | log
DANet R-101-D8 769x769 40000 12.8 1.07 V100 79.88 81.47 config model | log
DANet R-50-D8 512x1024 80000 - - V100 79.34 - config model | log
DANet R-101-D8 512x1024 80000 - - V100 80.41 - config model | log
DANet R-50-D8 769x769 80000 - - V100 79.27 80.96 config model | log
DANet R-101-D8 769x769 80000 - - V100 80.47 82.02 config model | log

ADE20K

Method Backbone Crop Size Lr schd Mem (GB) Inf time (fps) Device mIoU mIoU(ms+flip) config download
DANet R-50-D8 512x512 80000 11.5 21.20 V100 41.66 42.90 config model | log
DANet R-101-D8 512x512 80000 15 14.18 V100 43.64 45.19 config model | log
DANet R-50-D8 512x512 160000 - - V100 42.45 43.25 config model | log
DANet R-101-D8 512x512 160000 - - V100 44.17 45.02 config model | log

Pascal VOC 2012 + Aug

Method Backbone Crop Size Lr schd Mem (GB) Inf time (fps) Device mIoU mIoU(ms+flip) config download
DANet R-50-D8 512x512 20000 6.5 20.94 V100 74.45 75.69 config model | log
DANet R-101-D8 512x512 20000 9.9 13.76 V100 76.02 77.23 config model | log
DANet R-50-D8 512x512 40000 - - V100 76.37 77.29 config model | log
DANet R-101-D8 512x512 40000 - - V100 76.51 77.32 config model | log

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}
}