行人重识别
一个小超斌
计算机视觉方向在读
展开
-
Salience-Guided Cascaded Suppression Network for Person Re-identification阅读笔记
论文:Salience-Guided Cascaded Suppression Network for Person Re-identification来源:CVPR 2020论文地址动机:当前在全局特征和局部特征上结和注意力机制作为最后的特征表示已经成为了一种潮流,但在某些情况下,reid可能取决于在不同情况下显著特征所掩盖的不容易被注意到的特征。也就是说网络在学习到显著特征时会忽略学习其他重要的特征,而这些被显著特征掩盖起来的特征在不同场景中往往能起到重要的作用(比如身体、衣服甚至是鞋子)。解决原创 2020-10-18 13:38:18 · 512 阅读 · 0 评论 -
Hetero-Center Loss for Cross-Modality Person Re-Identification阅读笔记
论文题目:Hetero-Center Loss for Cross-Modality Person Re-Identification来源:Neurocomputing动机:对于跨模态reid,大多数的研究都是关注于提高类间的特征差异去解决问题(也就是提高不同ID行人图片之间的差异型),而很少关注于跨模态问题中类内的相似性(跨模态同ID图片之间的相似性)。因此,本文作者着力于减少cross-modality问题中类内部的变化。解决方法:1、Hetero-Center loss(HC loss)通过约束原创 2020-10-08 17:29:56 · 638 阅读 · 1 评论 -
Dynamic Dual-Attentive Aggregation Learning for Visible-Infrared Person Re-Identification阅读笔记
论文题目:Dynamic Dual-Attentive Aggregation Learning for Visible-Infrared Person Re-Identification来源:ECCV2020论文地址:https://arxiv.org/pdf/2007.09314.pdf动机:原创 2020-09-15 21:32:40 · 1496 阅读 · 1 评论 -
Person Re-identification with Deep Similarity-Guided Graph Neural Network阅读笔记
论文题目:Person Re-identification with Deep Similarity-Guided Graph Neural Network来源: ECCV2008论文地址: https://arxiv.org/abs/1807.09975动机: 这篇论文以往的行人重识别模型往往是给出一张probe图像,在gallery中找到相同id的行人图像,只关注probe和gallery其中一张图像之间的相似度关系,而忽略了不同probe-gallery图像对之间的关系。举个例子:给出一张pr原创 2020-09-04 17:17:34 · 319 阅读 · 0 评论 -
Cross-modality Person re-identification with Shared-Specific Feature Transfer笔记
论文:Cross-modality Person re-identification with Shared-Specific Feature Transfer来源:CVPR 2020地址:https://arxiv.org/abs/2002.12489动机:对于跨模态行人重识别任务,以往都只关注两个模态间的共享特征(share feature),而忽略了每个模态的Specific feature,作者通过利用每个模态的specific feature提高跨模态reid的性能。具体思路为:给定一红外线qu原创 2020-08-31 20:49:34 · 427 阅读 · 0 评论 -
High-Order Information Matters: Learning Relation and Topology for Occluded Person Re-Identificat阅读笔记
论文:High-Order Information Matters: Learning Relation and Topologyfor Occluded Person Re-Identification来源:CVPR 2020原创 2020-08-21 15:19:55 · 2128 阅读 · 0 评论