Research map / Label-efficient and robust learning / Recognition and long-tail
End2End Occluded Face Recognition by Masking Corrupted Features
Haibo Qiu, Dihong Gong, Zhifeng Li, Wei Liu, Dacheng Tao
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 |
|---|---|---|---|
| add occluded data augmentation (ablation) | Baseline · Occ-LFW-2.0 | verification accuracy (%): 82.05 -> 94.10 | Improved |
| add occluded data augmentation (ablation) | Baseline · MF1 | identification accuracy (%): 73.64 -> 72.50 | Hurt |
| add Mask Decoder (ablation) | Baseline-Aug · MF1 | identification accuracy (%): 72.50 -> 74.17 | Improved |
| add Occlusion Pattern Predictor (ablation) | Baseline-MD · Occ-MF1-AVG | identification accuracy (%): 68.19 -> 69.85 | Improved |
| add Occlusion Pattern Predictor (ablation) | Baseline-MD · MF1 | identification accuracy (%): 74.17 -> 75.27 | Improved |
| use FROM masks instead of no masks | Baseline · Occ-MF1-1.0 | identification accuracy (%): 32.00 -> 60.84 | Improved |
| increase grid K from 3 to 5 (ablation) | FROM · Occ-MF1 average | identification accuracy (%): 43.81 -> 44.83 | Improved |
| increase grid K from 5 to 6 (ablation) | FROM · Occ-MF1 average | identification accuracy (%): 44.83 -> 44.91 | No clear effect |
| add pattern prediction loss (lambda 0 to 1.0) (ablation) | FROM · Occ-MF1 average | identification accuracy (%): 43.48 -> 44.83 | Improved |
| lambda 0.5 instead of 1.0 (ablation) | FROM · Occ-MF1 average | identification accuracy (%): 45.00 vs 44.83 | No clear effect |
| apply masks on conv features instead of fc features (ablation) | FROM · Occ-MF1 average | identification accuracy (%): 44.05 -> 44.83 | Improved |
| use 3D masks instead of 2D masks (ablation) | FROM · Occ-MF1 average | identification accuracy (%): 44.35 -> 44.83 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 3 on most shared results.
- CosFace — CosFace: Large Margin Cosine Loss for Deep Face Recognition (CVPR 2018) · better on 6 of 7 shared results
- ArcFace — ArcFace: Additive Angular Margin Loss for Deep Face Recognition (CVPR 2019) · better on 1 of 2 shared results
- SphereFace — SphereFace: Deep Hypersphere Embedding for Face Recognition (CVPR 2017) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Occlusion Robust Face Recognition Based on Mask Learning With Pairwise Differential Siamese Network (ICCV 2019 · similarity 0.91)
- Occlusion-Free Face Alignment: Deep Regression Networks Coupled With De-Corrupt AutoEncoders (CVPR 2016 · similarity 0.88)
- Grid Loss: Detecting Occluded Faces (ECCV 2016 · similarity 0.87)
- MSML: Enhancing Occlusion-Robustness by Multi-Scale Segmentation-Based Mask Learning for Face Recognition (AAAI 2022 · similarity 0.87)
- Masked Face Recognition with Generative-to-Discriminative Representations (ICML 2024 · similarity 0.86)
- 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.
