Research map / Image generation / Text-to-image diffusion
Fair Generation without Unfair Distortions: Debiasing Text-to-Image Generation with Entanglement-Free Attention
Jeonghoon Park, Juyoung Lee, Chaeyeon Chung, Jaeseong Lee, Jaegul Choo, Jindong Gu
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 target-attribute regularization L_reg^trg (ablation) | EFA · WinoBias T_basic, Gender | PSNR: 27.76 -> 32.52 | Improved |
| add target-attribute regularization L_reg^trg (ablation) | EFA · WinoBias T_basic, Gender | DINO: 0.890 -> 0.916 | Improved |
| add target-attribute regularization L_reg^trg (ablation) | EFA · WinoBias T_basic, Gender | deviation ratio: 0.07 -> 0.06 | No clear effect |
| add human segmentation mask in training (ablation) | EFA · WinoBias T_basic, Gender | PSNR: 29.31 -> 32.52 | Improved |
| add human segmentation mask in training (ablation) | EFA · WinoBias T_basic, Gender | LPIPS: 0.0562 -> 0.0411 | Improved |
| add human segmentation mask in training (ablation) | EFA · WinoBias T_basic, Gender | deviation ratio: 0.04 -> 0.06 | Hurt |
| apply EFA only at 16x16 layers instead of adding 32x32 layers (ablation) | EFA · WinoBias T_basic, Gender | PSNR: 31.79 -> 32.52 | Improved |
| apply EFA only at 16x16 layers instead of adding 32x32 and 64x64 layers (ablation) | EFA · WinoBias T_basic, Gender | PSNR: 30.65 -> 32.52 | Improved |
| apply EFA at higher resolution layers (32x32, 64x64) (ablation) | EFA · WinoBias T_basic, Gender | deviation ratio: 0.06 -> 0.06 | No clear effect |
| replace Concept Algebra with Concept Algebra+ (target-attribute guidance) (ablation) | Concept Algebra · WinoBias T_basic, Gender | deviation ratio: 0.59 -> 0.02 | Improved |
| replace Concept Algebra with Concept Algebra+ (target-attribute guidance) (ablation) | Concept Algebra · WinoBias T_basic, Gender | PSNR: 21.10 -> 19.21 | Hurt |
| replace Concept Algebra with Concept Algebra+ (target-attribute guidance) (ablation) | Concept Algebra · COCO-no-person, Gender | FID: 2.41 -> 3.64 | Hurt |
Closest papers by meaning
- FairImagen: Post-Processing for Bias Mitigation in Text-to-Image Models (NeurIPS 2025 · similarity 0.90)
- Selectively Informative Description can Reduce Undesired Embedding Entanglements in Text-to-Image Personalization (CVPR 2024 · similarity 0.87)
- Rethinking Training for De-biasing Text-to-Image Generation: Unlocking the Potential of Stable Diffusion (CVPR 2025 · similarity 0.87)
- Fair Text-to-Image Diffusion via Fair Mapping (AAAI 2025 · similarity 0.87)
- Finetuning Text-to-Image Diffusion Models for Fairness (ICLR 2024 · similarity 0.87)
- Implicit Bias Injection Attacks against Text-to-Image Diffusion Models (CVPR 2025 · similarity 0.87)
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.
