Research map / Label-efficient and robust learning / Recognition and long-tail
Fair Facial Attribute Recognition via Group-Decoupled Vision Transformer with Mask-Guided Correlation Suppression
Huichang Huang, Kunchi Li, Si Chen, Da-Han Wang
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 Group-Attention (GD-ViT) to ViT (ablation) | Fair-GDMS (ViT-B/16) · CelebA (Male / Hair Color) | EOD: 0.2763 -> 0.1881 | Improved |
| add Group-Attention (GD-ViT) to ViT (ablation) | Fair-GDMS (ViT-B/16) · CelebA (Male / Hair Color) | accuracy (%): 90.2 -> 91.7 | Improved |
| add mask-guided correlation suppression on top of GD-ViT (ablation) | Fair-GDMS (ViT-B/16) · CelebA (Male / Hair Color) | EOD: 0.1881 -> 0.1173 | Improved |
| use learnable gate init 0.9 instead of 0.5 in GGDB (ablation) | Fair-GDMS (ViT-B/16) · CelebA (Male / Hair Color) | EOD: 0.1704 -> 0.1173 | Improved |
| use learnable gating instead of manual fusion weights (ablation) | Fair-GDMS (ViT-B/16) · CelebA (Male / Hair Color) | EOD: 0.1653 -> 0.1173 | Improved |
| set MIGF alpha to 0.5 instead of 0.1 or 0.7 (ablation) | Fair-GDMS (ViT-B/16) · CelebA (Male / Hair Color) | EOD: 0.1367 (0.1) / 0.1431 (0.7) -> 0.1173 | Improved |
| reassign majority samples to minority groups with prob 0.05 instead of 0.5 (ablation) | Fair-GDMS (ViT-B/16) · CelebA (Male / Hair Color) | accuracy (%): 81.3 -> 92.3 | Improved |
| reassign with prob 0.05 instead of 0.01 (ablation) | Fair-GDMS (ViT-B/16) · CelebA (Male / Hair Color) | accuracy (%): 91.5 -> 92.3 | Improved |
| add target-loss maximization in Stage 1 (ablation) | Fair-GDMS (ViT-B/16) · CelebA (Male / Hair Color) | masked regions | Improved |
| lower alpha from 0.5 to 0.2 (ablation) | Fair-GDMS (ViT-B/16) · CelebA (Male / Hair Color) | EOD: 0.1173 -> 0.1121 | No clear effect |
| use Fair-GDMS instead of Fair-ViT | Fair-GDMS · CelebA (Male / Attractive) | accuracy / fairness: Acc 84.0 -> 83.2; DP 0.2837 -> 0.1976 | Mixed |
Closest papers by meaning
- Robust RGB-D Face Recognition Using Attribute-Aware Loss (TPAMI 2019 · similarity 0.86)
- Learning Disentangled Representation for Fair Facial Attribute Classification via Fairness-aware Information Alignment (AAAI 2021 · similarity 0.86)
- Mitigating Demographic Bias in Facial Datasets with Style-Based Multi-attribute Transfer (IJCV 2021 · similarity 0.85)
- Mitigating Face Recognition Bias via Group Adaptive Classifier (CVPR 2021 · similarity 0.85)
- Distributionally Generative Augmentation for Fair Facial Attribute Classification (CVPR 2024 · similarity 0.84)
- FairDD: Fair Dataset Distillation (NeurIPS 2025 · 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.
