Research map / Graph learning / Graph algorithms and clustering
Learn To Cluster Faces via Pairwise Classification
Junfu Liu, Di Qiu, Pengfei Yan, Xiaolin Wei
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 weighted-neighbor features to the pair classifier input (ablation) | pairwise classifier · MS-Celeb-1M (584K) | classifier accuracy: 84.7 -> 95.2 | Improved |
| concatenate original features with weighted-neighbor features (combined) (ablation) | pairwise classifier · MS-Celeb-1M (584K) | classifier accuracy: 95.2 -> 95.6 | Improved |
| use combined features instead of original features (precision) (ablation) | pairwise classifier · MS-Celeb-1M (584K) | precision: 92.2 -> 95.4 | Improved |
| combined features instead of weighted-neighbor only (recall) (ablation) | pairwise classifier · MS-Celeb-1M (584K) | recall: 97.9 -> 97.5 | Hurt |
| add rank weighting to density (power 5 vs power 0) (ablation) | pair selection density · MS-Celeb-1M (584K) | Pairwise F-score: 88.57 -> 90.67 | Improved |
| add rank weighting to density (power 5 vs power 0) (ablation) | pair selection density · MS-Celeb-1M (584K) | BCubed F-score: 88.3 -> 89.54 | Improved |
| raise density power from 5 to 7 (ablation) | rank-weighted density · MS-Celeb-1M (584K) | Pairwise F-score: 90.67 -> 90.32 | Hurt |
| raise density power from 1 to 3 (ablation) | rank-weighted density · MS-Celeb-1M (584K) | Pairwise F-score: 89.69 -> 90.47 | Improved |
| replace graph-level GCN learning with pair-level classification | face clustering · MS-Celeb-1M (584K) | inference time: 11.5m -> 1.7m | Improved |
| replace GCN(V+E) with pairwise classification | face clustering · MS-Celeb-1M (5.21M) | Pairwise F-score: 79.3 -> 82.41 | Improved |
| replace GCN(V+E) with pairwise classification | face clustering · IJB-B (1845 ids) | Pairwise F-score: DANet 82.8 vs ours 82.7 | No clear effect |
| replace GCN(V+E) with pairwise classification | face clustering · DeepFashion | Pairwise F-score: 38.47 -> 37.67 | Hurt |
Closest papers by meaning
- Structure-Aware Face Clustering on a Large-Scale Graph With 107 Nodes (CVPR 2021 · similarity 0.92)
- Linkage Based Face Clustering via Graph Convolution Network (CVPR 2019 · similarity 0.91)
- Learning to Cluster Faces via Confidence and Connectivity Estimation (CVPR 2020 · similarity 0.90)
- STAR-FC: Structure-Aware Face Clustering on Ultra-Large-Scale Graphs (TPAMI 2023 · similarity 0.90)
- Learning to Cluster Faces on an Affinity Graph (CVPR 2019 · similarity 0.90)
- On Mitigating Hard Clusters for Face Clustering (ECCV 2022 · 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.
