Research map / Label-efficient and robust learning / Recognition and long-tail
Consensus-Driven Propagation in Massive Unlabeled Data for Face Recognition
Xiaohang Zhan, Ziwei Liu, Junjie Yan, Dahua Lin, Chen Change Loy
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 CDP with 10-fold unlabeled data | Tiny NASNet-A baseline · MegaFace | rank-1 identification rate: 61.78 -> 78.18 | Improved |
| add CDP with 10-fold unlabeled data | Tiny NASNet-A baseline · IJB-A | TPR@FPR=0.001: 75.87 -> 90.64 | Improved |
| use Inception-ResNet V2 as base model with CDP | CDP · MegaFace | rank-1 identification rate: 58.68 -> 81.88 | Improved |
| add CDP on top of ArcFace baseline | ArcFace with cleaner data · MegaFace | rank-1 identification rate: 76.93% -> 83.68% | Improved |
| add eight committee members to voting (ablation) | pair voting · validation set | pairwise recall: 0.680 -> 0.807 | Improved |
| add more committee members (4 to 8) to voting (ablation) | pair voting · validation set | pair selection precision: 0.987 -> 0.979 | Mixed |
| add affinity vector I_A to mediator input (ablation) | mediator · validation set | pairwise precision: 0.822 -> 0.888 | Improved |
| add neighbors distribution vector I_D to mediator input (ablation) | mediator · validation set | pairwise precision: 0.888 -> 0.912 | Improved |
| replace voting with mediator (ablation) | 8-member committee · validation set | pairwise precision: 0.876 -> 0.912 | Improved |
| use heterogeneous instead of homogeneous committee (ablation) | mediator · validation set | pairwise precision: 0.831 -> 0.912 | Improved |
| increase k in k-NN from 10 to 40 (ablation) | CDP · validation set | pairwise precision: 0.940 -> 0.874 | Hurt |
| increase k in k-NN from 10 to 40 (ablation) | CDP · validation set | pairwise recall: 0.810 -> 0.837 | Improved |
Closest papers by meaning
- Learning to Cluster Faces on an Affinity Graph (CVPR 2019 · similarity 0.87)
- Structure-Aware Face Clustering on a Large-Scale Graph With 107 Nodes (CVPR 2021 · similarity 0.86)
- From Face Recognition to Models of Identity: A Bayesian Approach to Learning about Unknown Identities from Unsupervised Data (ECCV 2018 · similarity 0.86)
- Vec2Face: Scaling Face Dataset Generation with Loosely Constrained Vectors (ICLR 2025 · similarity 0.85)
- GIF: Generative Inspiration for Face Recognition at Scale (CVPR 2025 · similarity 0.85)
- TopoFR: A Closer Look at Topology Alignment on Face Recognition (NeurIPS 2024 · similarity 0.84)
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.
