Research map / Vision-language and multimodal / CLIP and zero-shot
Target Bias Is All You Need: Zero-Shot Debiasing of Vision-Language Models with Bias Corpus
Taeuk Jang, Hoin Jung, Xiaoqian Wang
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 ZSDebias to CLIP ResNet-50 | CLIP RN50 · CelebA zero-shot classification | accuracy gap: 7.82 -> 1.68 | Improved |
| apply ZSDebias to CLIP ViT-L/14 | CLIP ViT-L/14 · CelebA zero-shot classification | accuracy gap: 12.24 -> 2.42 | Improved |
| apply ZSDebias to CLIP ResNet-50 | CLIP RN50 · CelebA zero-shot classification | worst-group accuracy: 72.49 -> 83.16 | Improved |
| apply ZSDebias to CLIP ViT-L/14 | CLIP ViT-L/14 · CelebA zero-shot classification | worst-group accuracy: 75.51 -> 78.18 | Improved |
| apply ZSDebias to CLIP ResNet-50 | CLIP RN50 · CelebA zero-shot classification | average accuracy: 80.31 -> 84.85 | Improved |
| apply ZSDebias to CLIP ViT-L/14 | CLIP ViT-L/14 · CelebA zero-shot classification | average accuracy: 87.75 -> 80.60 | Hurt |
| apply ZSDebias to CLIP ResNet-50 | CLIP RN50 · FairFace zero-shot retrieval | MaxSkew: 0.1478 -> 0.1238 | Improved |
| apply ZSDebias to CLIP ViT-L/14 | CLIP ViT-L/14 · FairFace zero-shot retrieval | MaxSkew: 0.1878 -> 0.1676 | Improved |
| use ZSDebias instead of CLIP-clip | CLIP ViT-L/14 · FairFace zero-shot retrieval | MaxSkew: 0.0334 -> 0.1676 | Hurt |
| use DBP instead of plain zero-shot | CLIP ViT-L/14 · FairFace zero-shot retrieval | MaxSkew: 0.1878 -> 0.2802 | Hurt |
| use ZSDebias instead of CLIP-clip | CLIP RN50 · CelebA zero-shot classification | accuracy gap: 4.01 -> 1.68 | Improved |
| use ZSDebias instead of ConAdapt | CLIP ViT-L/14 · CelebA zero-shot classification | worst-group accuracy: 89.44 -> 78.18 | Hurt |
Closest papers by meaning
- BendVLM: Test-Time Debiasing of Vision-Language Embeddings (NeurIPS 2024 · similarity 0.93)
- Joint Vision-Language Social Bias Removal for CLIP (CVPR 2025 · similarity 0.92)
- A Unified Debiasing Approach for Vision-Language Models across Modalities and Tasks (NeurIPS 2024 · similarity 0.91)
- Revisiting the Role of Language Priors in Vision-Language Models (ICML 2024 · similarity 0.90)
- The Neglected Tails in Vision-Language Models (CVPR 2024 · similarity 0.89)
- Enhancing Zero-Shot Vision Models by Label-Free Prompt Distribution Learning and Bias Correcting (NeurIPS 2024 · similarity 0.89)
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.
