Research map / Label-efficient and robust learning / Recognition and long-tail
Fair Contrastive Learning for Facial Attribute Classification
Check 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 |
|---|---|---|---|
| Replace SupCon with FSCL+, restricting negatives to same-sensitive-attribute, different-target examples and adding group-wise normalization | ResNet-18 facial attribute representation learning · CelebA (attractiveness; sensitive: male) | EO: 30.5 -> 6.5 | Helped |
| Add group-wise normalization to FSCL (ablation) | FSCL · CelebA (attractiveness; sensitive: male) | EO: 11.5 -> 6.5 | Helped |
| Add group-wise normalization to FSCL (ablation) | FSCL · CelebA (attractiveness; sensitive: male) | accuracy: 79.1 -> 79.1 | No clear effect |
| Add GRL adversarial training during representation learning (ablation) | FSCL+ · CelebA (attractiveness; sensitive: male) | EO: 6.5 -> 9.0 | Hurt |
| Add GRL adversarial training during representation learning (ablation) | SupCon · CelebA (attractiveness; sensitive: male) | accuracy: 80.5 -> 76.6 | Hurt |
| Add fair data augmentation (ablation) | FSCL+ · CelebA (attractiveness; sensitive: male) | EO: 6.5 -> 4.2 | Helped |
| Add fair data augmentation (ablation) | FSCL+ · CelebA (attractiveness; sensitive: male) | accuracy: 79.1 -> 79.6 | Helped |
| Add GRL during classifier training and unfreeze the encoder (ablation) | FSCL+ · CelebA (attractiveness; sensitive: male) | EO: 6.5 -> 20.5 | Hurt |
| Use FSCL without target class labels during representation learning (ablation) | FSCL† · CelebA (attractiveness; sensitive: male) | EO: 29.4 -> 14.8 | Helped |
| Use FSCL+ with pseudo-labels for unavailable sensitive labels (ablation) | FSCL+ · CelebA (attractiveness; sensitive: male; 1/2 sensitive labels) | EO: 13.4 -> 12.8 | Helped |
| Use FSCL+ with pseudo-labels for unavailable sensitive labels (ablation) | FSCL+ · CelebA (attractiveness; sensitive: male; 1/4 sensitive labels) | EO: 18.7 -> 13.4 | Helped |
| Use FSCL+ with pseudo-labels for unavailable sensitive labels (ablation) | FSCL+ · CelebA (attractiveness; sensitive: male; 1/10 sensitive labels) | EO: 20.7 -> 16.5 | Helped |
Closest papers by meaning
- Distributionally Generative Augmentation for Fair Facial Attribute Classification (CVPR 2024 · similarity 0.87)
- Kernel-based Equalized Odds: A Quantification of Accuracy-Fairness Trade-off in Fair Representation Learning (NeurIPS 2025 · similarity 0.87)
- Classes Are Not Equal: An Empirical Study on Image Recognition Fairness (CVPR 2024 · similarity 0.87)
- FairDD: Fair Dataset Distillation (NeurIPS 2025 · similarity 0.86)
- A Tale of Two Classes: Adapting Supervised Contrastive Learning to Binary Imbalanced Datasets (CVPR 2025 · similarity 0.85)
- Simple and Effective Specialized Representations for Fair Classifiers (NeurIPS 2025 · 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.