Research map / Label-efficient and robust learning / Recognition and long-tail
FAN-Face: a Simple Orthogonal Improvement to Deep Face Recognition
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 |
|---|---|---|---|
| add heatmap stacking at the input (ablation) | FAN-Face (ResNet-34) · IJB-B 1:1 verification (1M VGGFace2 subset) | TAR @ FAR=1e-4: 0.859 -> 0.864 | Helped |
| add FAN feature integration on top of heatmaps (ablation) | FAN-Face (ResNet-34) · IJB-B 1:1 verification (1M VGGFace2 subset) | TAR @ FAR=1e-4: 0.864 -> 0.876 | Helped |
| use high-to-low instead of low-to-high FAN features (ablation) | FAN-Face (ResNet-34) · IJB-B 1:1 verification (1M VGGFace2 subset) | TAR @ FAR=1e-4: 0.871 -> 0.876 | Helped |
| replace additive Simple layer with Basic Layer (norm + concat) (ablation) | FAN-Face (ResNet-34) · IJB-B 1:1 verification (1M VGGFace2 subset) | TAR @ FAR=1e-4: 0.857 -> 0.876 | Helped |
| use Advanced Layer instead of Basic Layer (ablation) | FAN-Face (ResNet-34) · IJB-B 1:1 verification (1M VGGFace2 subset) | TAR @ FAR=1e-4: 0.876 -> 0.873 | No clear effect |
| use 1-many instead of 1-1 connectivity (ablation) | FAN-Face (ResNet-34) · IJB-B 1:1 verification (1M VGGFace2 subset) | TAR @ FAR=1e-4: 0.876 -> 0.874 | No clear effect |
| use multi-task landmark head instead of feature integration (ablation) | FAN-Face (ResNet-34) · IJB-B 1:1 verification (1M VGGFace2 subset) | TAR @ FAR=1e-4: 0.876 -> 0.731 | Hurt |
| stack heatmaps into several FRN layers (H2+) vs input only (ablation) | FAN-Face (ResNet-34) · IJB-B 1:1 verification (1M VGGFace2 subset) | TAR @ FAR=1e-4: 0.864 -> 0.866 | No clear effect |
| use ArcFace loss instead of softmax with FAN-Face (ablation) | FAN-Face (ResNet-34) · IJB-B 1:1 verification (1M VGGFace2 subset) | TAR @ FAR=1e-4: 0.796 -> 0.876 | Helped |
| add FAN-Face integration to ArcFace ResNet-50 | FAN-Face (ResNet-50) · IJB-B (VGGFace2) TAR @ FAR=1e-4 | TAR: 0.902 -> 0.911 | Helped |
| add FAN-Face integration to ArcFace ResNet-34 | FAN-Face (ResNet-34) · IJB-C (VGGFace2) TAR @ FAR=1e-4 | TAR: 0.912 -> 0.926 | Helped |
| add FAN-Face integration to ArcFace ResNet-100 on MS1MV2 | FAN-Face (ResNet-100) · IJB-B (MS1MV2) | TAR (%): 94.5 -> 95.4 | Helped |
Closest papers by meaning
- Super-FAN: Integrated Facial Landmark Localization and Super-Resolution of Real-World Low Resolution Faces in Arbitrary Poses With GANs (CVPR 2018 · similarity 0.87)
- Joint Super-Resolution and Alignment of Tiny Faces (AAAI 2020 · similarity 0.87)
- LAFS: Landmark-based Facial Self-supervised Learning for Face Recognition (CVPR 2024 · similarity 0.87)
- AnchorFace: An Anchor-based Facial Landmark Detector Across Large Poses (AAAI 2021 · similarity 0.85)
- Generalizable Face Landmarking Guided by Conditional Face Warping (CVPR 2024 · similarity 0.84)
- Vec2Face: Scaling Face Dataset Generation with Loosely Constrained Vectors (ICLR 2025 · 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.