【汇总贴】人脸识别相关论文的优秀解读博客汇总

秒速五厘米 2022-05-23 00:05 186阅读 0赞

## 人脸识别相关论文的优秀解读博客汇总 ##

为了便于以后查阅，将看到的人脸识别相关论文的解读博客做一个汇总。

### 0、综述： ###

1.  [如何走近深度学习人脸识别？你需要这篇超长综述 | 附开源代码][Link 1]
2.  [关于度量学习的解释][Link 2]

### 一、FaceNet ###

**论文地址：**[https://arxiv.org/abs/1503.03832][https_arxiv.org_abs_1503.03832]

**参考博客：**  
1. [史上最全的FaceNet源码使用方法和讲解（附预训练模型下载）][FaceNet]  
2. [facenet 代码阅读笔记：如何训练基于triplet-loss的模型][facenet _triplet-loss]  
2. [谷歌人脸识别系统FaceNet解析][FaceNet 1]  
3. [FaceNet—深度学习与人脸识别的二次结合][FaceNet 2]  
4. [【人脸识别】FaceNet- A Unified Embedding for Face Recognition and Clustering][FaceNet- A Unified Embedding for Face Recognition and Clustering]

### 二、对于损失函数的改进 ###

##### 1、Deep Face Recognition #####

针对 Triplet Loss训练收敛缓慢的问题，提出了先用传统的 softmax粗训练人脸识别模型，再用Triplet Loss进行细致训练。  
**论文地址：**[http://www.robots.ox.ac.uk/~vgg/software/vgg\_face/][http_www.robots.ox.ac.uk_vgg_software_vgg_face]

**参考博客：**  
[人脸识别：Deep Face Recognition论文阅读][Deep Face Recognition]

##### 2、 In Defense of the Triplet Loss for Person Re-Identification #####

主要贡献是：① 设计了新的Triplet Loss（去掉了平方项），并和其它变种进行了对比；② 引进了 Batch Hard Sampling  
**论文地址：**[https://arxiv.org/abs/1703.07737][https_arxiv.org_abs_1703.07737]

**github地址：**[https://github.com/VisualComputingInstitute/triplet-reid][https_github.com_VisualComputingInstitute_triplet-reid]

**参考博客：**  
1. [Re-ID with Triplet Loss][]  
2. [【论文笔记】In Defense of the Triplet Loss for Person Re-Identification][In Defense of the Triplet Loss for Person Re-Identification]

##### 3、Sampling matters in deep embedding learning #####

主要贡献：① 改进了样本选取的方法（Distance Weighted Sampling Method）；② 提出了一种triplet loss的变体—— Margin Based Loss。据说这两个都比上一篇论文中的效果更好，但是没有亲测。

特别需要说明的是，这篇论文里回顾了contrastive loss,triplet loss（facenet论文）的损失函数和sample的选取方式，并说明了其相应的优缺点，可以说集合了这两篇论文的精华思想。看一篇，顶三篇。

**论文地址：**[https://arxiv.org/abs/1706.07567][https_arxiv.org_abs_1706.07567]

**参考博客：**  
1. [深度学习新的采样方式和损失函数论文笔记-知乎][-]  
2. [知乎-如何评价Sampling Matters in Deep Embedding Learning？  
][-_Sampling Matters in Deep Embedding Learning_]

[Link 1]: https://blog.csdn.net/c9Yv2cf9I06K2A9E/article/details/79825600
[Link 2]: https://blog.csdn.net/nehemiah_li/article/details/44230053
[https_arxiv.org_abs_1503.03832]: https://arxiv.org/abs/1503.03832
[FaceNet]: https://blog.csdn.net/u013044310/article/details/79556099
[facenet _triplet-loss]: http://www.mamicode.com/info-detail-2096766.html
[FaceNet 1]: https://zhuanlan.zhihu.com/p/24837264
[FaceNet 2]: http://www.cnblogs.com/xiaohuahua108/p/6505756.html
[FaceNet- A Unified Embedding for Face Recognition and Clustering]: https://zhuanlan.zhihu.com/p/29922142
[http_www.robots.ox.ac.uk_vgg_software_vgg_face]: http://www.robots.ox.ac.uk/~vgg/software/vgg_face/
[Deep Face Recognition]: http://yongyuan.name/blog/deep-face-recognition-note.html
[https_arxiv.org_abs_1703.07737]: https://arxiv.org/abs/1703.07737
[https_github.com_VisualComputingInstitute_triplet-reid]: https://github.com/VisualComputingInstitute/triplet-reid
[Re-ID with Triplet Loss]: https://blog.csdn.net/shuzfan/article/details/70069822
[In Defense of the Triplet Loss for Person Re-Identification]: https://blog.csdn.net/qq_21190081/article/details/78417215
[https_arxiv.org_abs_1706.07567]: https://arxiv.org/abs/1706.07567
[-]: https://zhuanlan.zhihu.com/p/27748177
[-_Sampling Matters in Deep Embedding Learning_]: https://www.zhihu.com/question/61748966/answer/212286125