Research map / Vision-language and multimodal / Multimodal LLMs
GenderBias-VL: Benchmarking Gender Bias in Vision Language Models via Counterfactual Probing
Yisong Xiao, Xianglong Liu, QianJia Cheng, Zhenfei Yin, Siyuan Liang, Jiapeng Li, Jing Shao, Aishan Liu, Dacheng Tao
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 |
|---|---|---|---|
| scale LLaVA1.5 LLM from 7B to 13B | LLaVA1.5 · GenderBias-VL VL-Bias | B_ovl: 1.85 -> 2.91 | Hurt |
| scale LLaVA1.5 LLM from 7B to 13B | LLaVA1.5 · GenderBias-VL VL-Bias | Acc: 52.15 -> 59.08 | Improved |
| upgrade LLaVA1.5-13B to LLaVA1.6-13B with larger training corpus | LLaVA · GenderBias-VL VL-Bias | B_ovl: 2.91 -> 3.29 | Hurt |
| upgrade LLaVA1.5-13B to LLaVA1.6-13B with larger training corpus | LLaVA · GenderBias-VL VL-Bias | Acc: 59.08 -> 67.70 | Improved |
| add RLHF to LLaVA1.5-13B (LLaVA-RLHF) | LLaVA1.5-13B · GenderBias-VL VL-Bias | B_ovl: 2.91 -> 4.15 | Hurt |
| add RLHF to LLaVA1.5-13B (LLaVA-RLHF) | LLaVA1.5-13B · GenderBias-VL VL-Bias | Acc: 59.08 -> 63.04 | Improved |
| compare commercial Gemini-Pro with GPT-4o on Top-10 biased pairs | Gemini-Pro · GenderBias-VL Top-10 pairs | B^o_ovl: nearly 3 times the bias level of GPT-4o | Hurt |
| swap option order | LLaVA1.5 models · GenderBias-VL | Delta Acc: 96.24 and 89.33 average Delta Acc for LLaVA1.5 and LAv2 | Hurt |
| use CLIP image encoder alone instead of the LVLM built on it | CLIP vs LVLMs · GenderBias-VL V-Bias | B^o_ovl: CLIP 9.11 vs LLaVA1.5-7B 0.85 | Mixed |
| compare visual vs language modality bias | LVLMs · GenderBias-VL | B_micro correlation: rho 0.77 and tau 0.60 | No clear effect |
Closest papers by meaning
- Revealing and Reducing Gender Biases in Vision and Language Assistants (VLAs) (ICLR 2025 · similarity 0.92)
- SocialCounterfactuals: Probing and Mitigating Intersectional Social Biases in Vision-Language Models with Counterfactual Examples (CVPR 2024 · similarity 0.90)
- Are Models Biased on Text without Gender-related Language? (ICLR 2024 · similarity 0.88)
- On Fairness of Unified Multimodal Large Language Model for Image Generation (NeurIPS 2025 · similarity 0.88)
- OpenBias: Open-set Bias Detection in Text-to-Image Generative Models (CVPR 2024 · similarity 0.87)
- Language Bias in LVLMs: From In-Depth Analysis to Simple and Effective Mitigation (ICML 2026 · 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.
