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

Table 2

Comparison of clustering results in the YouTube Faces (YTF) database.

Method	F-measure	FBcubed	Clusters
ResNet-29 + k-means ()	0.629	0.657	1595
ResNet-29 + k-means ()	0.595	0.656	1894
ResNet-29 + k-means ()	0.494	0.610	3050
ResNet-29 + GLC	0.832	0.787	21529
Approximate Rank-order [7]	0.71	—	—
ConPaC [4]	—	—	—
ResNet-29 + Approximate Rank-order	0.788	0.800	5563
ResNet-29 + ours (proposed)	0.889	0.854	3050

The true number of identities is 1,595, and the total number of face images is 621,126.