Research map / Label-efficient and robust learning / Recognition and long-tail
Do We Really Need to Collect Millions of Faces for Effective Face Recognition?
Iacopo Masi, Anh Tuấn Trần, Tal Hassner, Jatuporn Toy Leksut, Gérard Medioni
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 pose augmentation to training data (ablation) | VGGNet-19 on CASIA · IJB-A verification (with video pooling) | TAR @ FAR=0.01: 75.0 -> 86.3 | Improved |
| add pose augmentation to training data (ablation) | VGGNet-19 on CASIA · IJB-A identification (with video pooling) | Rank-1: 77.8 -> 88.0 | Improved |
| add shape variation to pose augmentation (ablation) | VGGNet-19 on CASIA · IJB-A verification (with video pooling) | TAR @ FAR=0.01: 86.3 -> 87.8 | Improved |
| add expression (closed mouth) synthesis to pose and shape augmentation (ablation) | VGGNet-19 on CASIA · IJB-A verification (with video pooling) | TAR @ FAR=0.01: 87.8 -> 88.1 | Improved |
| add expression synthesis to pose and shape augmentation (ablation) | VGGNet-19 on CASIA · IJB-A verification (without video pooling) | TAR @ FAR=0.001: 62.0 -> 63.6 | Improved |
| add video pooling of features within each video (ablation) | Full method · IJB-A verification | TAR @ FAR=0.001: 63.6 -> 71.0 | Improved |
| add video pooling of features within each video (ablation) | Full method · IJB-A identification | Rank-1: 87.2 -> 89.1 | Improved |
| replace mean fusion with SoftMax template fusion (ablation) | Template score pooling · IJB-A verification | TAR @ FAR=0.01: 79.9 -> 86.6 | Improved |
| replace max fusion with SoftMax template fusion (ablation) | Template score pooling · IJB-A verification | TAR @ FAR=0.001: 46.4 -> 63.6 | Improved |
| combine in-plane alignment with test-time pose synthesis (ablation) | Full method · IJB-A verification (without video pooling) | TAR @ FAR=0.01: 86.6 -> 87.8 | Improved |
| combine in-plane alignment with test-time pose synthesis (ablation) | Full method · IJB-A identification (with video pooling) | Rank-1: 89.1 -> 90.6 | Improved |
| use only rendered images instead of in-plane aligned images at test time (ablation) | Full method · IJB-A verification (with video pooling) | TAR @ FAR=0.01: 88.1 -> 84.8 | Hurt |
Closest papers by meaning
- BroadFace: Looking at Tens of Thousands of People at Once for Face Recognition (ECCV 2020 · similarity 0.89)
- The MegaFace Benchmark: 1 Million Faces for Recognition at Scale (CVPR 2016 · similarity 0.88)
- Vec2Face: Scaling Face Dataset Generation with Loosely Constrained Vectors (ICLR 2025 · similarity 0.87)
- How to Boost Face Recognition with StyleGAN? (ICCV 2023 · similarity 0.87)
- Pre-training Strategies and Datasets for Facial Representation Learning (ECCV 2022 · similarity 0.87)
- HyperFace: Generating Synthetic Face Recognition Datasets by Exploring Face Embedding Hypersphere (ICLR 2025 · similarity 0.86)
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.
