Research map / Label-efficient and robust learning / Recognition and long-tail
Pose-Invariant Face Recognition via Adaptive Angular Distillation
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 |
|---|---|---|---|
| distill frontal-center knowledge with KL loss (ablation) | ResNet-50 + ArcFace · CFP-FP | accuracy: 96.74 -> 96.85 | Helped |
| add PAD loss to the teacher-student method (ablation) | ResNet-50 + ArcFace + KL · CFP-FP | accuracy: 96.85 -> 97.78 | Helped |
| distill frontal-center knowledge with KL loss (ablation) | ResNet-50 + ArcFace · IJB-A | TAR @ FAR=0.001: 88.7 ± 1.5 -> 89.2 ± 1.8 | Helped |
| add PAD loss to the teacher-student method (ablation) | ResNet-50 + ArcFace + KL · IJB-A | TAR @ FAR=0.001: 89.2 ± 1.8 -> 91.5 ± 1.4 | Helped |
| use pose-adaptive sample weights instead of αn,m = 1 (ablation) | PAD · CFP-FP | accuracy: 97.44 -> 97.78 | Helped |
| use identity-adaptive margins instead of fixed margins (ablation) | PAD · CFP-FP | accuracy: 97.66 -> 97.78 | Helped |
| increase the number of selected negative frontal centers from 1 to 5 (ablation) | PAD · CFP-FP | accuracy: 97.61 -> 97.78 | Helped |
| increase the number of selected negative frontal centers from 5 to 15 (ablation) | PAD · CFP-FP | accuracy: 97.78 -> 97.58 | Hurt |
| add the teacher-student frontal-center method and PAD (ablation) | ResNet-50 + ArcFace · CFP-FF | accuracy: 99.72 -> 99.75 | No clear effect |
| add the teacher-student frontal-center method and PAD (ablation) | ResNet-50 + ArcFace · LFW | accuracy: 99.75 -> 99.77 | No clear effect |
| use ResNet-18 instead of ResNet-50 (ablation) | ArcFace + KL + PAD · CFP-FP | accuracy: 97.78 -> 96.14 | Hurt |
| use ResNet-34 instead of ResNet-50 (ablation) | ArcFace + KL + PAD · IJB-A | TAR @ FAR=0.001: 91.5 ± 1.4 -> 90.8 ± 1.5 | Hurt |
Closest papers by meaning
- Towards Pose Invariant Face Recognition in the Wild (CVPR 2018 · similarity 0.89)
- Discriminative Invariant Kernel Features: A Bells-and-Whistles-Free Approach to Unsupervised Face Recognition and Pose Estimation (CVPR 2016 · similarity 0.86)
- Disentangled Representation Learning GAN for Pose-Invariant Face Recognition (CVPR 2017 · similarity 0.86)
- Adversarial Pose Regression Network for Pose-Invariant Face Recognitions (AAAI 2021 · similarity 0.86)
- Pose-Disentangled Contrastive Learning for Self-Supervised Facial Representation (CVPR 2023 · similarity 0.85)
- Dual Pose-invariant Embeddings: Learning Category and Object-specific Discriminative Representations for Recognition and Retrieval (CVPR 2024 · similarity 0.85)
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.