Skip to content

Face Clustering Benchmarks

yl-1993 edited this page Jun 6, 2019 · 5 revisions

Dataset Overview

MS-1M-Celeb

Parts #class #instance
0 8573 576494
1 8573 584013
3 25719 1740301
5 42865 2890517
7 60011 4046365
9 77157 5206761

YouTube-Faces

Parts #class #instance
0 159 14653
1 1436 140629

Detailed Results

MS-1M-Celeb

Method #cluster Precision Recall F-score
Approx Rank Order (knn=80, th=0) 286098 99.77 7.2 13.42
MiniBatchKmeans (ncluster=5000, bs=100) 5000 45.48 80.98 58.25
KNN DBSCAN (knn=80, th=0.7, eps=0.7, min=40) 290847 62.38 50.66 55.92
FastHAC (dist=0.72, single) 122754 92.07 57.28 70.63
CDP (single model, th=0.7) 83034 80.19 70.47 75.02
GCN-D (0.7, 0.75) 100102 95.41 67.79 79.26
GCN-D (0.65, 0.7, 0.75) 79072 94.64 71.53 81.48
GCN-D (0.6, 0.65, 0.7, 0.75) 74919 94.60 72.52 82.10

YouTube-Faces

Method #cluster Precision Recall F-score
Approx Rank Order (knn=80, th=0) 17677 99.95 45.2 62.25
MiniBatchKmeans (ncluster=1500, bs=100) 1500 76.87 51.86 61.93
MiniBatchKmeans (ncluster=1500, bs=16) 1500 81.97 97.67 67.71
KNN DBSCAN (knn=80, th=0.7, eps=0.7, min=40) 36400 98.99 79.46 88.16
KNN DBSCAN (knn=80, th=0.7, eps=0.7, min=1) 2841 98.06 84.82 90.96
FastHAC (dist=0.7, single) 3621 99.64 87.31 93.07
CDP (single model, th=0.7, knn=160, maxsz=1600) 2736 94.59 91.73 93.13
Clone this wiki locally