「Computer Vision」Notes on Vortex Pooling

QQ Group: 428014259
Sina Weibo：小锋子Shawn
Tencent E-mail：403568338@qq.com
http://blog.youkuaiyun.com/dgyuanshaofeng/article/details/82941912

作者：Chen-Wei Xie, Hong-Yu Zhou, Jianxin Wu
单位：National Key Laboratory for Novel Software Technology, Nangjing University

0 摘要

观点是"when predicting the category of a given pixel, the regions chose to the target are more important than those far from it"。提出Vortex Pooling进行解决。与DeepLab v3相比，在VOC 2012验证集上，提升1.5%，在VOC 2012测试集上，提升0.6%，用Vortex Pooling替换ASPP模块。计算代价，用Vortex Pooling的系统为10.13FPS，而DeepLab v3为10.37FPS。从这里看出，VP比ASPP计算更复杂，或者VP是建立在ASPP上。

1 介绍

2 相关工作

3 提出方法

3.1 DeepLab v3回顾

略

3.2 讨论和分析

3.1 模块A

[1] Vortex Pooling Improving Context Representation in Semantic Segmentation 2018 [paper]