Effective and Generalizable Graph-Based Clustering for Faces in the Wild

Table 1

Comparison of clustering results in the Labeled Faces in the Wild (LFW) database.

Method	F-measure	FBcubed	Clusters
ResNet-29 + k-means ()	0.158	0.750	5749
ResNet-29 + k-means ()	0.153	0.749	5761
ResNet-29 + k-means ()	0.143	0.749	6352
ResNet-29 + GLC	0.920	0.911	6809
Approximate Rank-order [7]	0.870	—	6508
ConPaC [4]	0.965	0.922	6352
ResNet-29 + Approximate Rank-order	0.696	0.859	6564
ResNet-29 + ours (proposed)	0.973	0.934	5761

The true number of identities is 5,749, and the total number of face images is 13,233.