Research map / Label-efficient and robust learning / Recognition and long-tail
Towards Interpretable Face Recognition
Bangjie Yin, Luan Tran, Haoxiang Li, Xiaohui Shen, Xiaoming Liu
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 |
|---|---|---|---|
| use FAD threshold t=260 instead of forcing all elements equal (t=320) (ablation) | CASIA-Net with SAD+FAD · IJB-A | verification TAR @FAR=.01 (%): 74.6 -> 79.2 | Improved |
| use FAD threshold t=260 instead of t=130 (ablation) | CASIA-Net with SAD+FAD · IJB-A | verification TAR @FAR=.01 (%): 79.0 -> 79.2 | No clear effect |
| add FAD loss to SAD loss (ablation) | CASIA-Net · IJB-A manual occlusion | verification TAR @FAR=.01 (%): 66.6 -> 76.1 | Improved |
| add FAD loss to SAD loss (ablation) | CASIA-Net · IJB-A manual occlusion | identification Rank-1 (%): 81.2 -> 88.0 | Improved |
| add FAD loss to SAD loss (ablation) | CASIA-Net · IJB-A | verification TAR @FAR=.01 (%): 78.1 -> 79.2 | Improved |
| add SAD loss to FAD loss (ablation) | CASIA-Net · IJB-A | verification TAR @FAR=.01 (%): 76.7 -> 79.2 | Improved |
| add SAD loss to FAD loss (ablation) | CASIA-Net · IJB-A manual occlusion | identification Rank-1 (%): 85.1 -> 88.0 | Improved |
| replace black static window with Gaussian noise dynamic-size occlusion in FAD (ablation) | CASIA-Net · IJB-A | identification Rank-1 (%): 89.4 -> 89.9 | Improved |
| add large magnitude filtering with d=95.83% (ablation) | CASIA-Net · peak location std (pos./neg.) | std of peak locations: 25.7/25.7 -> 12.9/13.4 | Improved |
| add SAD loss to base CNN (ablation) | CASIA-Net · spreadness of average peak locations | spreadness d-bar: 6.9 -> 17.1 | Improved |
| add FAD loss to SAD only (ablation) | CASIA-Net · spreadness of average peak locations | spreadness d-bar: 17.1 -> 18.7 | Improved |
| train baseline CASIA-Net with synthetic occlusion augmentation | CASIA-Net · IJB-A | verification TAR @FAR=.01 (%): 74.3 -> 78.9 | Improved |
Closest papers by meaning
- Network Dissection: Quantifying Interpretability of Deep Visual Representations (CVPR 2017 · similarity 0.88)
- Masked Face Recognition with Generative-to-Discriminative Representations (ICML 2024 · similarity 0.86)
- PICNN: A Pathway towards Interpretable Convolutional Neural Networks (AAAI 2024 · similarity 0.85)
- Enhancing Pre-trained Representation Classifiability can Boost its Interpretability (ICLR 2025 · similarity 0.85)
- Measuring Per-Unit Interpretability at Scale Without Humans (NeurIPS 2024 · similarity 0.85)
- A Convolutional Neural Network Interpretable Framework for Human Ventral Visual Pathway Representation (AAAI 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.
