Research map / Vision-language and multimodal / Human and embodied modeling
Mitigating Test-Time Bias for Fair Image Retrieval
Fanjie Kong, Shuai Yuan, Weituo Hao, Ricardo Henao
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 |
|---|---|---|---|
| apply PBM test-time debiasing with full gender labels | CLIP ViT-B/16 · Occupation 1 gender | AbsBias@100: .6231 -> .0000 | Improved |
| apply PBM with full labels | CLIP ViT-B/16 · Occupation 1 gender | mAP@100: 65.0 -> 54.5 | Hurt |
| apply PBM with full labels | CLIP ViT-B/16 · Occupation 1 gender | Recall@100: 58.3 -> 49.1 | Hurt |
| use PBM with pre-trained classifier instead of CLIP-clip | CLIP ViT-B/16 · Occupation 1 gender | AbsBias@100: .5048 -> .1404 | Improved |
| predict gender with pre-trained MLP classifier instead of zero-shot embedding | PBM · Occupation 1 gender | AbsBias@100: 0.09688 -> .1404 | Mixed |
| use zero-shot prompt instead of zero-shot embedding for attribute prediction | PBM · Occupation 1 gender | AbsBias@100: 0.09688 -> .0560 | Improved |
| use adversarial training for debiasing | CLIP ViT-B/16 · Occupation 1 gender | mAP@100: 65.0 -> 49.7 | Hurt |
| use Debias Prompt | CLIP ViT-B/16 · Occupation 1 gender | AbsBias@100: .6231 -> .6373 | No clear effect |
| apply PBM with pre-trained classifier | CLIP ViT-B/16 · Occupation 2 race | AbsBias@100: .5002 -> .09552 | Improved |
| apply PBM | CLIP ViT-B/16 · COCO-1k | Bias@1: .0900 -> .0402 | Improved |
| apply PBM | CLIP ViT-B/16 · COCO-1k | Recall@1: 48.2 -> 37.3 | Hurt |
| apply PBM | CLIP ViT-B/16 · Flickr30K | Bias@1: .1150 -> .0360 | Improved |
Closest papers by meaning
- FairImagen: Post-Processing for Bias Mitigation in Text-to-Image Models (NeurIPS 2025 · similarity 0.90)
- FairRAG: Fair Human Generation via Fair Retrieval Augmentation (CVPR 2024 · similarity 0.86)
- BendVLM: Test-Time Debiasing of Vision-Language Embeddings (NeurIPS 2024 · similarity 0.86)
- Finetuning Text-to-Image Diffusion Models for Fairness (ICLR 2024 · similarity 0.86)
- Classifier-to-Bias: Toward Unsupervised Automatic Bias Detection for Visual Classifiers (CVPR 2025 · similarity 0.86)
- Fair Text-to-Image Diffusion via Fair Mapping (AAAI 2025 · similarity 0.85)
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.
