Research map / Label-efficient and robust learning / Recognition and long-tail
Dynamic Feature Learning for Partial Face Recognition
Lingxiao He, Haiqing Li, Qi Zhang, Zhenan Sun
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 DFM instead of MR-CNN for partial-face matching | Partial face recognition · Partial-LFW, 1,000 classes | Rank-1 accuracy: 24.7% -> 27.3% | Improved |
| Use DFM instead of the strongest listed baseline, MR-CNN | Partial face recognition · CASIA-NIR-Distance | Rank-1 accuracy: 85.97 -> 94.96 | Improved |
| Use DFM instead of the strongest listed baseline, MR-CNN | Partial face recognition · Partial-YTF | Rank-1 accuracy: 57.00 -> 61.00 | Improved |
| Use DFM with spatial feature maps rather than resize partial images to fixed-size inputs | Partial face recognition · Partial-LFW, 1,000 classes | Rank-1 accuracy: 20.9% -> 27.3% | Improved |
| Use DFM instead of VGGFace on arbitrary partial-face probes | Partial face verification · LFW verification protocol | True positive rate | Improved |
| Use holistic probes rather than a mixture of partial and holistic probes | DFM · Partial-LFW, 1,000 classes | Rank-1 accuracy: 41.9% -> 56.8% | Improved |
| Use a mixture of partial and holistic probes rather than only partial probes | DFM · Partial-LFW, 1,000 classes | Rank-1 accuracy: 27.3% -> 41.9% | Improved |
| Present the upper face area rather than the lower face area | DFM · Partial-LFW | Rank-1 accuracy: 7.8% -> 39.2% | Improved |
| Increase the cropped patch size | DFM · Partial-LFW | Rank-1 accuracy | Improved |
| Set similarity-guided constraint weight α to 2.1 and sparsity weight β to 0.1 (ablation) | DFM · Partial-LFW, 1,000 classes | Rank-1 accuracy: 27.3% | Improved |
| Use larger α after its low-to-high sweep (ablation) | DFM · Partial-LFW | Rank-1 accuracy | Mixed |
| Set sparsity weight β to 0.1 rather than the other tested values (ablation) | DFM · Partial-LFW | Rank-1 accuracy | Improved |
Closest papers by meaning
- Killing Two Birds With One Stone: Efficient and Robust Training of Face Recognition CNNs by Partial FC (CVPR 2022 · similarity 0.86)
- BroadFace: Looking at Tens of Thousands of People at Once for Face Recognition (ECCV 2020 · similarity 0.85)
- Context-Aware Local Binary Feature Learning for Face Recognition (TPAMI 2017 · similarity 0.84)
- ArcFace: Additive Angular Margin Loss for Deep Face Recognition (CVPR 2019 · similarity 0.84)
- Neural Aggregation Network for Video Face Recognition (CVPR 2017 · similarity 0.83)
- ProxyFusion: Face Feature Aggregation Through Sparse Experts (NeurIPS 2024 · 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.
