人体解析--Look into Person: Self-supervised Structure-sensitive Learning

Look into Person: Self-supervised Structure-sensitive Learning and A New Benchmark for Human Parsing
CVPR2017

https://arxiv.org/abs/1703.05446
LIP benchmark:http://hcp.sysu.edu.cn/lip/index.php

针对 Human parsing 这个问题,这里提出了一个新的 benchmark: Look into Person (LIP),随后对比几个经典的human parsing 算法,分析优缺点,随后我们提出自己的算法 :self-supervised structure-sensitive learning approach ,将 human pose structures 引入到 人体解析中来。

首先来看看 Look into Person (LIP) 数据库:
和其他两个数据库的对比:
人体解析--Look into Person: Self-supervised Structure-sensitive Learning_第1张图片

当前几个数据库的规模:
人体解析--Look into Person: Self-supervised Structure-sensitive Learning_第2张图片

LIP中19个人体部件的图像标记数量:
人体解析--Look into Person: Self-supervised Structure-sensitive Learning_第3张图片

不同情况的人体图像数量
人体解析--Look into Person: Self-supervised Structure-sensitive Learning_第4张图片

几个经典算法在 LIP上的效果对比:
人体解析--Look into Person: Self-supervised Structure-sensitive Learning_第5张图片

不同图像条件下的对比:
人体解析--Look into Person: Self-supervised Structure-sensitive Learning_第6张图片

人体解析--Look into Person: Self-supervised Structure-sensitive Learning_第7张图片

本文提出的算法,引入 人体结构 信息:
人体解析--Look into Person: Self-supervised Structure-sensitive Learning_第8张图片

人体解析--Look into Person: Self-supervised Structure-sensitive Learning_第9张图片

PASCAL-Person-Part dataset 效果对比:
人体解析--Look into Person: Self-supervised Structure-sensitive Learning_第10张图片

人体解析--Look into Person: Self-supervised Structure-sensitive Learning_第11张图片

你可能感兴趣的:(CVPR2017,人)