Research map / Graph learning / Graph algorithms and clustering
Learning to Cluster Faces via Confidence and Connectivity Estimation
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add the connectivity estimator GCN-E to GCN-V (ablation) | Ours (V) · MS-Celeb-1M 584K | F_P: 87.14 -> 87.93 | Helped |
| add the connectivity estimator GCN-E to GCN-V (ablation) | Ours (V) · DeepFashion | F_P: 33.07 -> 38.47 | Helped |
| use supervised neighborhood confidence instead of unsupervised density (ablation) | vertex confidence estimator · MS1M-584K | F_P: 81.78 -> 82.76 | Helped |
| define confidence from neighborhood similarity rather than same-class average similarity (ablation) | vertex confidence estimator · DeepFashion | F_P: 30.11 -> 32.24 | Helped |
| rebuild the affinity graph using the top GCN embedding (ablation) | vertex confidence estimator · MS1M-584K | F_P: 82.76 -> 87.14 | Helped |
| apply GCN-E to only the high-confidence vertices rather than all vertices (ablation) | Ours (V + E) · DeepFashion | F_P | Mixed |
| remove the higher-confidence candidate-set restriction for GCN-E (ablation) | Ours (V + E) · DeepFashion | F_P: 38.47 -> 29.85 | Hurt |
| connect each vertex to two neighbors instead of one (ablation) | Ours (V + E) · DeepFashion | F_P: 38.47 -> 1.19 | Hurt |
| connect each vertex to three neighbors instead of one (ablation) | Ours (V + E) · DeepFashion | F_P: 38.47 -> 0.31 | Hurt |
| use the single-neighbor tree partition rather than multi-neighbor links (ablation) | Ours (V + E) · DeepFashion | F_B: 56.43 -> 60.06 | Helped |
| use GCN-E on top of GCN-V (ablation) | Ours · MS-Celeb-1M 5.21M | Time: 4.5m -> 11.5m | Hurt |
| use confidence-based vertex estimation and connectivity-based clustering instead of heuristic supervised clustering | Ours (V + E) · MS-Celeb-1M 5.21M | F_P: 77.87 -> 79.3 | Helped |
Closest papers by meaning
- FC-DGCN: Deep Graph Convolutional Network for Face Clustering and Recognition (NeurIPS 2026 · similarity 0.93)
- Learning to Cluster Faces on an Affinity Graph (CVPR 2019 · similarity 0.93)
- Linkage Based Face Clustering via Graph Convolution Network (CVPR 2019 · similarity 0.93)
- Structure-Aware Face Clustering on a Large-Scale Graph With 107 Nodes (CVPR 2021 · similarity 0.92)
- Ada-NETS: Face Clustering via Adaptive Neighbour Discovery in the Structure Space (ICLR 2022 · similarity 0.90)
- Density-Aware Feature Embedding for Face Clustering (CVPR 2020 · similarity 0.89)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.