Research map / Image generation / Text-to-image diffusion
TIBET: Identifying and Evaluating Biases in Text-to-Image Generative Models
Aditya Chinchure, Pushkar Shukla, Gaurav Bhatt, Kiri Salij, Kartik Hosanagar, Leonid Sigal, Matthew Turk
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 |
|---|---|---|---|
| use VQA concept-based CAS instead of CLIP embedding similarity (ablation) | TIBET image comparison · User Study 2 (societal biases) | ranking correlation with human ratings: +0.07 -> +0.51 | Improved |
| use VQA concept-based CAS instead of CLIP embedding similarity (ablation) | TIBET image comparison · User Study 2 (societal biases) | Top-1 accuracy: 50% -> 75% | Improved |
| use CLIP text embeddings in bipartite matching of concepts (ablation) | TIBET image comparison · User Study 2 | ranking correlation: -0.08 vs +0.51 for CAS | Hurt |
| use normalized MAD instead of Wasserstein distance or standard deviation as bias measure (ablation) | TIBET · Sensitivity to 18% VQA error | relative score change: 28.65% (std) / 15.15% (Wasserstein) -> 13.11% (MAD) | Improved |
| VQA errors propagate into CAS (ablation) | TIBET · Sensitivity analysis, 30 prompts | relative CAS change: 18% VQA error -> 4.73% CAS change | No clear effect |
| upgrade from Stable Diffusion 1.5 to 2.1 | text-to-image model · 11 occupation prompts | mean CAS gap male vs female: 0.56/0.44 -> 0.52/0.48 | Improved |
| apply ITI-GEN using TIBET counterfactual images as reference sets | Stable Diffusion · 11 occupation prompts | CAS gap male vs female | Improved |
Closest papers by meaning
- OpenBias: Open-set Bias Detection in Text-to-Image Generative Models (CVPR 2024 · similarity 0.88)
- Diverse Yet Biased: Towards Mitigating Biases in Generative AI (Student Abstract) (AAAI 2024 · similarity 0.88)
- FairImagen: Post-Processing for Bias Mitigation in Text-to-Image Models (NeurIPS 2025 · similarity 0.87)
- Who Evaluates the Evaluations? Objectively Scoring Text-to-Image Prompt Coherence Metrics with T2IScoreScore (TS2) (NeurIPS 2024 · similarity 0.87)
- Multi-Group Proportional Representations for Text-to-Image Models (CVPR 2025 · similarity 0.87)
- See It from My Perspective: How Language Affects Cultural Bias in Image Understanding (ICLR 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.
