Learning to Cluster Faces on an Affinity Graph (LTC)

Paper

Download the pretrained models in the model zoo.

Test cluster detection

sh scripts/dsgcn/test_cluster_det_ms1m.sh

Test cluster segmentation

# predict iop and then conduct seg
sh scripts/dsgcn/test_cluster_det_iop_ms1m.sh
sh scripts/dsgcn/test_cluster_seg_ms1m.sh

[Optional] GCN-D Upper Bound It yields the performance when accuracy of GCN-D is 100%.

sh scripts/dsgcn/step_by_step/gcn_d_upper_bound.sh

Train cluster detection

sh scripts/dsgcn/train_cluster_det_ms1m.sh

Train cluster segmentation

# seg training uses the ground-truth iop
sh scripts/dsgcn/train_cluster_seg_ms1m.sh

Users can choose different proposals in dsgcn/configs or design your own proposals for training and testing.

Generally, more proposals leads to better results. You can control the number of proposals to strike a balance between time and performance.