Research map / Label-efficient and robust learning / OOD and anomaly detection
Improving Training and Inference of Face Recognition Models via Random Temperature Scaling
Read the paperCheck 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 RTS instead of the strongest reported baseline | Face recognition models trained on DeepGlint · CFP-FP | verification accuracy: 98.08 -> 98.09 | No clear effect |
| Use RTS instead of the strongest reported baseline | Face recognition models trained on DeepGlint · LFW | verification accuracy: 99.82 -> 99.77 | Hurt |
| Use RTS for OOD detection instead of MagFace | ResNet64 face recognition model · OOD test set | AUC: 97.48 -> 98.38 | Helped |
| Train with 20% Gaussian-blurred images using RTS instead of the baseline | Face recognition model trained on CASIA-WebFace · LFW | verification accuracy: 99.40 -> 99.45 | Helped |
| Set KL divergence weight λ to 0.1 rather than 10 (ablation) | RTS · LFW (CASIA-WebFace) | verification accuracy: 99.43 -> 98.43 | Hurt |
| Set KL divergence weight λ to 1 rather than 10 (ablation) | RTS · LFW (CASIA-WebFace) | verification accuracy: 99.43 -> 99.23 | Hurt |
| Set degree of freedom δ to 8 rather than 16 (ablation) | RTS · LFW (CASIA-WebFace) | verification accuracy: 99.43 -> 99.30 | Hurt |
| Set degree of freedom δ to 32 rather than 16 (ablation) | RTS · LFW (CASIA-WebFace) | verification accuracy: 99.43 -> 99.33 | Hurt |
| Use a dynamic ArcFace margin during training | ArcFace-based methods, including RTS · Verification benchmarks and OOD detection | performance | No clear effect |
Closest papers by meaning
- Data Uncertainty Learning in Face Recognition (CVPR 2020 · similarity 0.90)
- Adaptive Uncertainty Estimation via High-Dimensional Testing on Latent Representations (NeurIPS 2023 · similarity 0.86)
- On the Limitations of Temperature Scaling for Distributions with Overlaps (ICLR 2024 · similarity 0.85)
- SURE: SUrvey REcipes for building reliable and robust deep networks (CVPR 2024 · similarity 0.84)
- Robust Visual Recognition with Class-Imbalanced Open-World Noisy Data (AAAI 2024 · similarity 0.83)
- RepFace: Refining Closed-Set Noise with Progressive Label Correction for Face Recognition (AAAI 2025 · similarity 0.83)
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.