Research map / Label-efficient and robust learning / Noisy labels
Noise-Tolerant Paradigm for Training Face Recognition CNNs
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 |
|---|---|---|---|
| Apply the proposed theta-distribution sample-weighting paradigm | ResNet-64 with ArcFace trained on original MS-Celeb-1M · LFW | accuracy: 99.21 -> 99.72 | Helped |
| Apply the proposed theta-distribution sample-weighting paradigm | ResNet-64 with ArcFace trained on original MS-Celeb-1M · AgeDB-30 | accuracy: 90.85 -> 96.70 | Helped |
| Apply the proposed theta-distribution sample-weighting paradigm | ResNet-64 with ArcFace trained on original MS-Celeb-1M · CFP-FP | accuracy: 93.38 -> 96.40 | Helped |
| Apply the proposed theta-distribution sample-weighting paradigm | ResNet-64 with ArcFace trained on original MS-Celeb-1M · YTF | accuracy: 95.64 -> 97.36 | Helped |
| Apply the proposed theta-distribution sample-weighting paradigm | ResNet-64 with ArcFace trained on original MS-Celeb-1M · MegaFace Challenge 1 | accuracy: 70.16 -> 78.69 | Helped |
| Use the sample weight in the AM-Loss scaling parameter rather than multiplying the loss (ablation) | Proposed paradigm with L2-Softmax, 40% noise · LFW | accuracy: 89.91 -> 93.90 | Helped |
| Use the sample weight in the AM-Loss scaling parameter rather than multiplying the loss (ablation) | Proposed paradigm with L2-Softmax, 40% noise · AgeDB(30) | accuracy: CNN_m1 is only evaluated on LFW | Helped |
| Use the sample weight in the AM-Loss scaling parameter rather than multiplying the loss (ablation) | Proposed paradigm with ArcFace, 40% noise · LFW | accuracy: 95.88 -> 97.23 | Helped |
| Use the sample weight in the AM-Loss scaling parameter rather than multiplying the loss (ablation) | Proposed paradigm with ArcFace, 40% noise · AgeDB(30) | accuracy: CNN_m1 is only evaluated on LFW | Helped |
| Train on noisy samples without sample weighting | ResNet-20 with L2-Softmax, WebFace-All at 40% noise · LFW | accuracy: 92.71 -> 85.63 | Hurt |
| Train on noisy samples without sample weighting | ResNet-20 with ArcFace, WebFace-All at 40% noise · LFW | accuracy: 96.53 -> 92.33 | Hurt |
| Emphasize semi-hard clean samples in the final training stage through the fused weighting strategies | CNN_m2 trained on WebFace-All · LFW, 0% noise, L2-Softmax | accuracy: 94.65 -> 96.28 | Helped |
Closest papers by meaning
- Unequal-Training for Deep Face Recognition With Long-Tailed Noisy Data (CVPR 2019 · similarity 0.89)
- Confidence-based Reliable Learning under Dual Noises (NeurIPS 2022 · similarity 0.88)
- Coresets for Robust Training of Deep Neural Networks against Noisy Labels (NeurIPS 2020 · similarity 0.88)
- Enhancing Noise-Robust Losses for Large-Scale Noisy Data Learning (AAAI 2025 · similarity 0.86)
- RepFace: Refining Closed-Set Noise with Progressive Label Correction for Face Recognition (AAAI 2025 · similarity 0.86)
- $\epsilon$-Softmax: Approximating One-Hot Vectors for Mitigating Label Noise (NeurIPS 2024 · 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.