Research map / Vision-language and multimodal / Human and embodied modeling
Quantifying Societal Bias Amplification in Image Captioning
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 the gender equalizer loss to NIC+ | NIC+ · MSCOCO gender | LIC: 7.2 -> 11.8 | Hurt |
| Add the gender equalizer loss to NIC+ | NIC+ · MSCOCO gender | Error: 12.9 -> 7.7 | Helped |
| Add the gender equalizer loss to NIC+ | NIC+ · MSCOCO race | LIC: 7.6 -> 7.2 | No clear effect |
| Mask gender-correlated objects in images (ablation) | SAT · MSCOCO gender | LIC_M: 44.4 -> 42.9 | Helped |
| Mask gender-correlated objects in images (ablation) | OSCAR · MSCOCO gender | LIC_M: 48.5 -> 46.2 | Helped |
| Mask people in images (ablation) | SAT · MSCOCO gender | LIC_M: 44.4 -> 39.1 | Helped |
| Mask people in images (ablation) | OSCAR · MSCOCO gender | LIC_M: 48.5 -> 39.7 | Helped |
| Mask both gender-correlated objects and people in images (ablation) | SAT · MSCOCO gender | LIC_M: 44.4 -> 37.2 | Helped |
| Mask both gender-correlated objects and people in images (ablation) | OSCAR · MSCOCO gender | LIC_M: 48.5 -> 39.0 | Helped |
| Use a larger caption vocabulary | image captioning models · MSCOCO gender | LIC | Helped |
| Improve caption BLEU-4 score | image captioning models · MSCOCO gender | LIC | Hurt |
Closest papers by meaning
- SocialCounterfactuals: Probing and Mitigating Intersectional Social Biases in Vision-Language Models with Counterfactual Examples (CVPR 2024 · similarity 0.88)
- FairImagen: Post-Processing for Bias Mitigation in Text-to-Image Models (NeurIPS 2025 · similarity 0.87)
- Joint Vision-Language Social Bias Removal for CLIP (CVPR 2025 · similarity 0.86)
- SANER: Annotation-free Societal Attribute Neutralizer for Debiasing CLIP (ICLR 2025 · similarity 0.86)
- Finetuning Text-to-Image Diffusion Models for Fairness (ICLR 2024 · similarity 0.86)
- OASIS Uncovers: High-Quality T2I Models, Same Old Stereotypes (ICLR 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.