yl-1993 / learn-to-cluster

Learning to Cluster Faces (CVPR 2019, CVPR 2020)

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Figure 3 in Learning to Cluster Faces via Confidence and Connectivity Estimation

longbowzhang opened this issue · comments

Hi @yl-1993 Nice work!
But I have one question regarding the Fig. 3 of the paper Learning to Cluster Faces via Confidence and Connectivity Estimation.
image
I am wondering why there are three red dots in "ours". If I understand correctly, one is GCN-V, and the other is GCN-V + GCN-E. Then what's the 3rd one?

Hi @longbowzhang , really sorry for the outdated reply!

From left to right, the three red dots are GCN-V, GCN-V+GCN-E(with small ρ), GCN-V + GCN-E(ρ=10%), where ρ is the portion of vertices using GCN-E as defined in Sec.3.3 .

To be more specific, the time(x-axis) for three red dots is 270s, 413s and 687s, and the pairwise f-score is 87.14, 87.23 and 87.55 respectively. The first and the last value are reported in last two rows of Table.1. For the middle value, we give a more detailed ablation study of ρ in Figure.5.