人脸聚类论文--Clustering Millions of Faces by Identity(Approximate rank-order clustering)

RT

前言

  1. 据说这是在工业界广泛使用的算法,经历过实践的考验
  2. 论文指导老师是人脸/聚类领域的巨佬, 学术新坑开启人:Anil K. Jain

浅析

总结来说:论文就是基于RO1做了少量改进, 提升了表现与性能。 (注意区分ARO, RO, ARO是改进版,也是这篇文章的主角)

ARO距离计算原理

建议先看看参考链接,解释的很清晰
文字版描述如下:

  1. 计算rank order
  2. 计算rank order distance : (d(A->B), d(B->A))
  3. 计算normalized pair distance between A & B, 利用step2 计算出来的距离

ARO改进1:

RO的算法流程:
在这里插入图片描述
上图的分析如下:
1.每张人脸N作为独立的分类, 构建N*N对pair(这里会有很多重复的pair(Ci, Cj)(Cj, Ci), 实现上可以去除 ),
迭代运行 step2 & step3:
2.然后计算pair之间距离得到candidate pair(同类的pair),
3.然后对candidate pairs进行merging. (每一个query 假设都有pairs对candidate pair, 时间复杂度O(pairs^2), 两次for循环. 过程中不断merge为新的cluster , 再去计算新的cluster之间的距离, 为什么要这么做? )

这里step3 合并成新的cluster之后, 作者定义了cluster level distance , 即cluster之间的距离(与单独两张个体图片的距离定义不同,大致如下图所示) 其实没有必要去定义/计算新的cluster 之间的距离, 直接算一遍step2, step3, 复杂度从O(C*N^2) 降为O(N^2)
(C is clusters)
在这里插入图片描述

ARO改进2:

1.rank order distance 公式(1.1采用了topk, 而不是遍历了所有的 neighbors 1.2 采用了Indicator (Ib)操作,只在乎是否存在, 而不是用rank_value)

图中1)公式为RO论文提出, 5)公式为ARO论文提出
在这里插入图片描述

效果与性能分析(TODO)

参考

参考链接

  • [1] zhu2011A Rank-Order Distance based Clustering Algorithm for Face Tagging
  • 0
    点赞
  • 2
    收藏
    觉得还不错? 一键收藏
  • 0
    评论
Abstract—Clustering face images according to their latent identity has two important applications: (i) grouping a collection of face images when no external labels are associated with images, and (ii) indexing for efficient large scale face retrieval. The clustering problem is composed of two key parts: representation and similarity metric for face images, and choice of the partition algorithm. We first propose a representation based on ResNet, which has been shown to perform very well in image classification problems. Given this representation, we design a clustering algorithm, Conditional Pairwise Clustering (ConPaC), which directly estimates the adjacency matrix only based on the similarities between face images. This allows a dynamic selection of number of clusters and retains pairwise similarities between faces. ConPaC formulates the clustering problem as a Conditional Random Field (CRF) model and uses Loopy Belief Propagation to find an approximate solution for maximizing the posterior probability of the adjacency matrix. Experimental results on two benchmark face datasets (LFW and IJB-B) show that ConPaC outperforms well known clustering algorithms such as k-means, spectral clustering and approximate Rank-order. Additionally, our algorithm can naturally incorporate pairwise constraints to work in a semi-supervised way that leads to improved clustering performance. We also propose an k-NN variant of ConPaC, which has a linear time complexity given a k-NN graph, suitable for large datasets. Index Terms—face clustering, face representation, Conditional Random Fields, pairwise constraints, semi-supervised clustering.

“相关推荐”对你有帮助么?

  • 非常没帮助
  • 没帮助
  • 一般
  • 有帮助
  • 非常有帮助
提交
评论
添加红包

请填写红包祝福语或标题

红包个数最小为10个

红包金额最低5元

当前余额3.43前往充值 >
需支付:10.00
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明:

1.余额是钱包充值的虚拟货币,按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载,可以购买VIP、付费专栏及课程。

余额充值