Research map / Image generation / Text-to-image diffusion
HoloFair: Unified T2I Fairness Evaluation and Fair-GRPO Debiasing
Ruyi Chen, Lu Zhou, Xiaogang Xu, Chiyu Zhang, Jiafei Wu, Liming Fang
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 Fair-GRPO debiasing | SD3.5M · HoloFair Eval | MGBI: 0.5211 -> 0.6772 | Improved |
| add Fair-GRPO debiasing | SD1.5 · HoloFair Eval | MGBI: 0.6554 -> 0.7881 | Improved |
| add Fair-GRPO debiasing | SD3.5M · HoloFair Eval | FID: 143.26 -> 135.09 | Improved |
| add Fair-GRPO debiasing | SD3.5M · HoloFair Eval | CLIP-Score: 0.2288 -> 0.2317 | Improved |
| add Fair-GRPO debiasing | SD3.5M · HoloFair Eval | Asr: 5.7919 -> 5.76 | No clear effect |
| add Fair-GRPO debiasing | SD1.5 · HoloFair Eval | Pickscore: 0.2079 -> 0.2071 | Hurt |
| fine-tune SD3.5M on RBD dataset instead of Fair-GRPO | SD3.5M · HoloFair Eval | MGBI: 0.5211 -> 0.6190 | Mixed |
| add frequency-domain input to DINO classifier (ablation) | DINO + F.T. · FCT | Overall accuracy: 79.67 -> 85.33 | Improved |
| add adaptive weight fusion to DINO + Fre. (ablation) | SpaFreq · FCT | Overall accuracy: 85.33 -> 89.67 | Improved |
| train classifier on mixed RBD instead of FairFace only (ablation) | SpaFreq · FCT | Race accuracy: 84.79 -> 92.28 | Improved |
| add gender reward alone (ablation) | Fair-GRPO · SD3.5M | MGBI: 0.5211 -> 0.6302 | Improved |
| add age and race rewards to gender reward (joint) (ablation) | Fair-GRPO · SD3.5M | MGBI: 0.6302 -> 0.6772 | Improved |
Closest papers by meaning
- FairImagen: Post-Processing for Bias Mitigation in Text-to-Image Models (NeurIPS 2025 · similarity 0.88)
- Finetuning Text-to-Image Diffusion Models for Fairness (ICLR 2024 · similarity 0.86)
- Stay Fair! Ensuring Group Fairness in Diffusion Models Across Guidance Scales (NeurIPS 2026 · similarity 0.86)
- Balancing Act: Distribution-Guided Debiasing in Diffusion Models (CVPR 2024 · similarity 0.86)
- Fair Deepfake Detectors Can Generalize (NeurIPS 2025 · similarity 0.86)
- AIM-Fair: Advancing Algorithmic Fairness via Selectively Fine-Tuning Biased Models with Contextual Synthetic Data (CVPR 2025 · similarity 0.86)
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.
