论文笔记：Deeply-Learned Part-Aligned Representations for Person Re-Identification

简介

这是ICCV2017的一篇person Re-id的论文。论文提出了一种新的对其方法，以便在摄像机移动的场景下，更准确地匹配不同姿势、位置的person re-identification。不同于以往常用的空间分割方法，本文采用人类身体部分分割（human body part partition）。

方法

以往的align方法主要是通过空间位置进行对其，但是检测框是算法生成的而非人为标注的，并且人的姿势也会改变，通过这种方法进行对其往往有巨大的误差，如图所示：

首先用一个FCN提取整张图片的特征图，对于这个特征图，使用K个分支网络提取区域特征，最后将这K个分支拼接起来就是最终的hunman representation。

其中，Feature map是一个3维的特征图，用t(x,y,c)表示第c个（x,y）位置的响应；分支网络学习到的结果是一个2维的mask——MkM_kMk（类似于像素级别的注意力权重），m_k(x,y)表示（x，y）位置落在第k个区域的程度。将t和m点乘，并通过一个平均池化，就可以得到一个分支（区域）的特征向量。
tk(x,y,c)=t(x,y,c)×mk(x,y)t_{k}(x,\ y,\ c)=t(x,\ y,\ c)\times m_{k}(x,\ y)tk(x, y, c)=t(x, y, c)×mk(x, y)
fk(c)=Averagex,y[tk(x,y,c)]{f}_{k}(c)=\mathrm{Average}_{x,y}[t_{k}(x,\ y,\ c)]fk(c)=Averagex,y[tk(x, y, c)]
计算得到的k个f通过线性降维再拼接，就可以得到整张图的特征向量。

如图表示了每个分支所提取的区域的可视化结果。由图可以看出基本上每个分支提取的都是人体相同的部分。

论文笔记：Deeply-Learned Part-Aligned Representations for Person Re-Identification相关推荐

论文笔记--Efficient Estimation of Word Representations in Vector Space
论文笔记--Efficient Estimation of Word Representations in Vector Space 1. 文章简介 2. 文章概括 3 文章重点技术 3.1 NNLM ...
【论文笔记——DIM】Learning Deep Representations By Mutual Information Estimation and Maximization
arxiv:https://arxiv.org/abs/1808.06670 code:https://github.com/rdevon/DIM report video:https://www.y ...
Visual Question Answering with Textual Representations for Images 论文笔记
Visual Question Answering with Textual Representations for Images 论文笔记一.Abstract 二.引言三.方法 3.1 Lang ...
论文笔记--Deep contextualized word representations
论文笔记--Deep contextualized word representations 1. 文章简介 2. 文章概括 3 文章重点技术 3.1 BiLM(Bidirectional Langu ...
论文笔记：Decoding Brain Representations by Multimodal Learning of Neural Activity and Visual Features
论文笔记:Decoding Brain Representations by Multimodal Learning of Neural Activity and Visual Features(通过 ...
论文笔记：Evolving Losses for Unsupervised Video Representation Learning
Evolving Losses for Unsupervised Video Representation Learning 论文笔记 Distillation Knowledge Distillat ...
【论文笔记】Encoding cloth manipulations using a graph of states and transitions
[论文笔记]Encoding cloth manipulations using a graph of states and transitions Abstract 问题: Cloth manipu ...
论文笔记-Vanilla Transformer：Character-Level Language Modeling with Deeper Self-Attention
论文笔记-Vanilla Transformer:Character-Level Language Modeling with Deeper Self-Attention 1. 介绍 2. Chara ...
《Improved Crowd Counting Method Based onScale-Adaptive Convolutional Neural Network》论文笔记
<Improved Crowd Counting Method Based onScale-Adaptive Convolutional Neural Network>论文笔记论文地址 ...

论文笔记：Deeply-Learned Part-Aligned Representations for Person Re-Identification

论文笔记：Deeply-Learned Part-Aligned Representations for Person Re-Identification

简介

方法

论文笔记：Deeply-Learned Part-Aligned Representations for Person Re-Identification相关推荐

最新文章

热门文章