
论文笔记REID
从天而降小可爱
这个作者很懒,什么都没留下…
展开
专栏收录文章
- 默认排序
- 最新发布
- 最早发布
- 最多阅读
- 最少阅读
-
CVPR2019 paper list for Person Re-ID
Perceive Where to Focus: Learning Visibility-Aware Part-Level Features for Partial Person Re-IdentificationYifan Sun, Qin Xu, Yali Li, Chi Zhang, Yikang Li, Shengjin Wang, Jian Sun[pdf] [bibtex]Inv...原创 2020-02-20 21:34:41 · 719 阅读 · 0 评论 -
ICCV2019 paper list Person Re-ID
Instance-Guided Context Rendering for Cross-Domain Person Re-Identification Yanbei Chen, Xiatian Zhu, Shaogang Gong [pdf] [supp] [bibtex]Mixed High-Order Attention Network for Person Re-Identificatio...原创 2020-02-20 21:14:26 · 913 阅读 · 0 评论 -
For Cross-Domain person Re-ID 文献速览
Instance-Guided Context Rendering for Cross-Domain Person Re-IdentificationIdea:使用U-net网络,生成源域行人图像和目标域背景相结合的多张图像,扩充数据集的延展性。Loss:使用了四种loss帮助网络进行训练,Adversarial Loss,Camera Loss,Context Loss,Identity...原创 2020-02-20 14:15:05 · 277 阅读 · 0 评论 -
AAAI2020 Person ReID paper List
Relation Network for Person Re-identificationHyunjong Park (Yonsei University); Bumsub Ham (Yonsei University)*Cross-Modality Paired-Images Generation for RGB-Infrared Person Re-IdentificationGuan...原创 2020-02-14 15:06:52 · 1920 阅读 · 0 评论 -
Bag of Tricks and A Strong Baseline for Deep Person Re-identification
相当于是一种强力有效的训练Re-ID的方式,基于全局特征,不需要多分支网络。主要训练如下:我们使用ImageNet下训练的参数来初始化ResNet50,将全连接参数改为N。N表示训练数据集中目标的个数。我们随机抽取P个对象,每个对象K张图来构成训练batch。最终的Batchsize为B=P×K。本文中,我们设置P=16,K=4。将每张样本resize成256×128像素大小,...原创 2020-02-13 11:15:10 · 214 阅读 · 0 评论 -
Perceive Where to Focus: Learning Visibility-aware Part-level Features for Partial Person Re-ID
局部可见感知模型——VPM论文摘要论文中考虑了行人重新识别 (re-ID) 的一个现实问题,即局部行人重识别问题。在局部行人重识别场景下,需要探索数据库图像中包含整体人物的比例大小。如果直接将局部行人图像与整体行人图像进行比较,则极端空间错位问题将显著地损害特征表征学习的判别能力。针对这个问题,本文提出了一个局部可见感知模型 (VPM),通过自监督学习感知可见区域内的特征,来解决局部行人重识别...原创 2019-05-03 11:30:24 · 1537 阅读 · 3 评论 -
可变形卷积解读和代码实现(TF)
转载:https://blog.csdn.net/mykeylock/article/details/77746499可变形卷积是指卷积核在每一个元素上额外增加了一个参数方向参数,这样卷积核就能在训练过程中扩展到很大的范围。可变形卷积的论文为:Deformable Convolutional Networks【1】而之前google一篇论文对这篇论文有指导意义:Spatial Tran...转载 2018-09-15 17:27:32 · 14920 阅读 · 12 评论 -
CVPR2018:TFusion论文解读
转载:https://blog.csdn.net/u011453773/article/details/79638489 CVPR2018: Unsupervised Cross-dataset Person Re-identification by Transfer Learning of Spatio-temporal Patterns论文可以在...转载 2018-09-03 08:21:30 · 656 阅读 · 0 评论 -
Re-ID论文分析(更新中)
【论文阅读】Diversity Regularized Spatiotemporal Attention for Video-based Person Re-identification - cyHungry的文章 - 知乎 https://zhuanlan.zhihu.com/p/35460367原创 2018-08-26 11:30:57 · 1289 阅读 · 0 评论 -
Deep Spatial Feature Reconstruction for Partial Person Re-identification: Freestyle Approach(CV18)
转载:https://blog.csdn.net/VacantField/article/details/79801925这是CVPR 2018上的论文,论文针对局部行人再识别提出一种更高效率高准确率的方法,中文直译深度空域特征重建,Deep Spatial Feature 摘要: 局部行人再识别是一个很有现实意义的挑战。所谓局部是指,在无约束环境中,行人容易被遮挡,有姿势和视角变化,所...转载 2018-08-20 21:12:42 · 1227 阅读 · 0 评论 -
Learning Discriminative Features with Multiple Granularities for Person Re-Identification
刷新三数据集纪录的跨镜追踪(行人再识别-ReID)技术云从科技在跨镜追踪(行人再识别)技术(ReID)上获取重大突破。同时在Market-1501,CUHK03,DukeMTMC-reID三个数据集刷新了世界纪录,其中最高在Market-1501上的首位命中率(Rank-1 Accuracy)达到96.6%,让跨镜追踪(ReID)在准确率上首次达到商用水平,人工智...转载 2018-08-22 10:12:20 · 1350 阅读 · 0 评论 -
【论文笔记】Margin Sample Mining Loss: A Deep Learning Based Method for Person Re-identification
摘要Person re-identification (ReID) is an important task in computer vision. Recently, deep learning with a metric learning loss has become a common framework for ReID. In this paper,...转载 2018-08-18 14:55:53 · 592 阅读 · 0 评论