花椒蛋糕-优快云博客

原创基于视频的行人重识别结果可视化

基于视频的行人重识别结果可视化基于视频的行人重识别结果可视化数据集读取test结果保存结果读取gallery排序结果可视化结果结束语基于视频的行人重识别结果可视化在网上搜集了很多帖子，想要白嫖可视化代码。但是发现并没有人分享，是视频重识别太小众了还单帧重识别太受宠了！！！！为此在郑哲东的基础上改出了基于视频的行人重识别结果可视化。郑哲东大佬的代码先放在这里：https://github.com/layumi/Person_reID_baseline_pytorch单帧可视化的代码就是里面的demo

2022-01-04 16:06:36 4289 10

原创行人重识别阅读笔记之Cross-modality Person re-identification with Shared-Specific Feature Transfer

行人重识别阅读笔记之Cross-modality Person re-identification with Shared-Specific Feature Transfer摘要介绍模型Two-stream feature extractorShared-Specific Transfer NetworkShared and specific complementary learningOptimization论文地址:Cross-modality Person re-identification wit

2020-12-08 18:16:14 934

原创行人重识别阅读笔记之Multi-Granularity Reference-Aided Attentive Feature Aggregation

行人重识别阅读笔记之Multi-Granularity Reference-Aided Attentive Feature Aggregation摘要介绍模型概述Reference-aided Attentive Feature AggregationMulti-Granularity AttentionLoss Design论文地址：https://arxiv.org/pdf/2003.12224.pdf摘要帧间存在冗余、新显示的外观、遮挡和运动模糊。提出多粒度参考辅助注意力特征聚合Multi-G

2020-12-08 17:32:33 709

原创行人重识别阅读笔记之Spatial-Temporal Graph Convolutional Network for Video-based Person Re-identification

行人重识别阅读笔记之Spatial-Temporal Graph Convolutional Network for Video-based Person Re-identification摘要介绍模型PreliminaryPatch Graph ConstructionTemporal GCN ModuleStructural GCN ModuleModel and Loss Functions论文连接：Spatial-Temporal Graph Convolutional Network for V

2020-12-08 17:13:51 2217 3

原创行人重识别阅读笔记之Salience-Guided Cascaded Suppression Network for Person Re-identification

行人重识别阅读笔记之Salience-Guided Cascaded Suppression Network for Person Re-identification介绍网络结构残差双重注意力模型Residual Dual Attention ModuleChannel-wise AttentionResidual Spatial Attention非局部多阶段特征融合Non-local Multistage Feature Fusion显著特征提取装置Salient Feature Extraction

2020-11-13 20:41:52 1192

原创行人重识别阅读笔记之Unity Style Transfer for Person Re-Identification

行人重识别阅读笔记之Unity Style Transfer for Person Re-Identification摘要主要问题CamStyle的不足UnityGAN论文贡献模型UnityGANUnityStyleDeep Re-ID ModelPipelineTrainingTesting总结摘要风格变换一直是ReID的一个主要挑战，其目的是在不同的摄像机下匹配相同的行人。现有的研究试图用相机不变描述子空间学习来解决这个问题。当不同相机拍摄的图像差异较大时，会产生更多的图像伪影。为了解决这个问，作

2020-11-07 17:37:40 1337 5

原创行人重识别之RGB-Infrared Cross-Modality Person Re-identification via Joint Pixel and Feature Alignment

行人重识别之RGB-Infrared Cross-Modality Person Re-identification via Joint Pixel and Feature Alignment简介主要问题模型结构1、像素对齐模块 Pixel Alignment Module2、特征对齐模块 Feature Alignment Module3、联合判别模块 Joint Discriminator Module简介paper：https://arxiv.org/abs/1910.05839本文作者关注到

2020-11-01 23:12:09 691 2

weixin_51138870的博客