Research map / Label-efficient and robust learning / Domain adaptation and generalization
Learning Meta Face Recognition in Unseen Domains
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 |
|---|---|---|---|
| Use domain-level meta-optimization with hard-pair, soft-classification, and domain-alignment losses | MFR · GFR-R I-IV | VR at FAR=0.01%: 64.65 -> 71.4; 55.4 -> 60.8; 60.86 -> 62.56; 78.6 -> 81.1 | Helped |
| Add hard-pair attention loss (ablation) | MFR · GFR-R I | VR at FAR=0.01%: 69.2 -> 71.4 | Helped |
| Add soft-classification loss (ablation) | MFR · GFR-R I | VR at FAR=0.01%: 69 -> 71.4 | Helped |
| Add domain-alignment loss (ablation) | MFR · GFR-R I | VR at FAR=0.01%: 70.8 -> 71.4 | Helped |
| Add meta-gradient computation (ablation) | MFR · GFR-R I | VR at FAR=0.01%: 68.35 -> 71.4 | Helped |
| Replace high-order meta-gradient with first-order approximation (ablation) | MFR · GFR-R I | VR at FAR=0.01%: 71.4 -> 70.9 | Hurt |
| Use first-order rather than high-order meta-gradient computation (ablation) | MFR · GFR-R | training time: 63% | Helped |
| Set the meta-train step size α to 0, removing meta-gradient computation (ablation) | MFR · GFR-R I-IV | VR at FAR=0.01%: 71.4 -> 68.35; 60.8 -> 57.75; 62.56 -> 60.7; 81.1 -> 79.6 | Hurt |
| Use meta-train and meta-test weighting γ=0.5 (ablation) | MFR · GFR-R I | performance | Helped |
| Sample two domains for meta-train and one for meta-test (ablation) | MFR · GFR-R I | performance | Helped |
| Fine-tune the base model on source domains with a randomly initialized classification template | Base · GFR-R | VR at FAR=0.01%: 64.65 -> 21.05; 55.4 -> 3.5; 60.86 -> 7.67; 78.6 -> 18.9 | Hurt |
| Jointly train on the base dataset and source domains | Base · GFR-R | VR at FAR=0.01%: 64.65 -> 65.3; 55.4 -> 56.3; 60.86 -> 59.41; 78.6 -> 79.1 | Mixed |
Closest papers by meaning
- Open Domain Generalization with Domain-Augmented Meta-Learning (CVPR 2021 · similarity 0.87)
- Cross-Domain Similarity Learning for Face Recognition in Unseen Domains (CVPR 2021 · similarity 0.87)
- HyperFace: Generating Synthetic Face Recognition Datasets by Exploring Face Embedding Hypersphere (ICLR 2025 · similarity 0.85)
- UIFace: Unleashing Inherent Model Capabilities to Enhance Intra-Class Diversity in Synthetic Face Recognition (ICLR 2025 · similarity 0.84)
- UMFN: Unified Multi-Domain Face Normalization for Joint Cross-domain Prototype Learning and Heterogeneous Face Recognition (CVPR 2025 · similarity 0.84)
- Test-Time Domain Generalization for Face Anti-Spoofing (CVPR 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.