Research map / Vision-language and multimodal / Human and embodied modeling
VisoGender: A dataset for benchmarking gender bias in image-text pronoun resolution
Siobhan Mackenzie Hall, Fernanda Gonçalves Abrantes, Hanwen Zhu, Grace Sodunke, Aleksandar Shtedritski, Hannah Rose Kirk
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 captioning models instead of vision-language encoders for pronoun resolution | VLMs on VisoGender · VisoGender resolution | resolution accuracy: CLIP 0.75; BLIP-2 0.84; GIT 0.84 | Improved |
| Resolve pronouns in scenes with two people of different perceived genders rather than single-person scenes | VLMs on VisoGender · VisoGender resolution | resolution accuracy: CLIP: 0.92 -> 0.36; BLIP-2: 0.92 -> 0.60 | Hurt |
| Use gender-neutral occupational queries for image retrieval | Vision-language encoders on VisoGender · VisoGender retrieval, two-person images | Bias@5: 0.11; 0.10; 0.17; 0.06; 0.11; 0.01 | Mixed |
| Reverse occupation and participant order in the prompt (ablation) | CLIP on two-person pronoun resolution · VisoGender resolution | resolution accuracy | Hurt |
| Reverse occupation and participant order in the retrieval template (ablation) | SLIP on VisoGender retrieval · VisoGender retrieval | Bias@5: 0.06 -> -0.18 | Mixed |
| Include a neutral pronoun option (ablation) | CLIP on VisoGender · VisoGender neutral pronoun resolution | neutral pronoun resolution rate: 0.17 overall; 0.31 for two people with different perceived genders | No clear effect |
| Include a neutral pronoun option (ablation) | BLIP-2 on VisoGender · VisoGender neutral pronoun resolution | neutral pronoun resolution rate: 0.01 overall; 0.00 for two people with different perceived genders | No clear effect |
Closest papers by meaning
- GenderBias-VL: Benchmarking Gender Bias in Vision Language Models via Counterfactual Probing (IJCV 2025 · similarity 0.91)
- Revealing and Reducing Gender Biases in Vision and Language Assistants (VLAs) (ICLR 2025 · similarity 0.90)
- Are Models Biased on Text without Gender-related Language? (ICLR 2024 · similarity 0.87)
- Auto-Search and Refinement: An Automated Framework for Gender Bias Mitigation in Large Language Models (NeurIPS 2025 · similarity 0.86)
- On Fairness of Unified Multimodal Large Language Model for Image Generation (NeurIPS 2025 · similarity 0.86)
- SocialCounterfactuals: Probing and Mitigating Intersectional Social Biases in Vision-Language Models with Counterfactual Examples (CVPR 2024 · 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.
