Research map / Vision-language and multimodal / Multimodal LLMs
MAVias: Mitigate any Visual Bias
Ioannis Sarridis, Christos Koutlis, Symeon Papadopoulos, Christos Diou
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 |
|---|---|---|---|
| add logit alignment term (ablation) | MAVias · Waterbirds (closed-set) | WG Acc.: 89.4 -> 93.7 | Improved |
| add logit alignment term (ablation) | MAVias · Waterbirds (closed-set) | Avg. Acc.: 95.2 -> 94.5 | Hurt |
| use OpenCLIP instead of BERT-L for tag embeddings (ablation) | MAVias · Waterbirds (closed-set) | WG Acc.: 91.7 -> 93.7 | Improved |
| use OpenCLIP instead of SigLip-L-16-256 (ablation) | MAVias · Waterbirds (closed-set) | WG Acc.: 92.7 -> 93.7 | Improved |
| use GPT-4o instead of Qwen2.5 7B as tagging LLM (ablation) | MAVias · Waterbirds (closed-set) | WG Acc.: 91.9 -> 93.7 | Improved |
| use GPT-4o instead of Llama3.1 8B (ablation) | MAVias · Waterbirds (closed-set) | WG Acc.: 93.6 -> 93.7 | No clear effect |
| embed all tags collectively instead of separately (ablation) | MAVias · Waterbirds (closed-set) | WG Acc.: 93.0 -> 93.7 | Improved |
| use full tag vocabulary instead of 10% (ablation) | MAVias · Waterbirds (closed-set) | WG Acc.: 78.1 -> 93.7 | Improved |
| use 30% vocabulary instead of 100% (ablation) | MAVias · Waterbirds (closed-set) | WG Acc.: 91.1 -> 93.7 | No clear effect |
| apply MAVias instead of best BU baseline (open-set) | MAVias · CelebA open-set | WG Acc.: 31.5 -> 66.7 | Improved |
| apply MAVias instead of best BU baseline (open-set) | MAVias · UrbanCars open-set | WG Acc.: 69.0 -> 84.4 | Improved |
| apply MAVias on ImageNet9 ORIGINAL test set | MAVias · ImageNet9 | ORIGINAL acc.: 97.69 -> 96.92 | Hurt |
Closest papers by meaning
- Discovering and Mitigating Visual Biases through Keyword Explanation (CVPR 2024 · similarity 0.88)
- Classifier-to-Bias: Toward Unsupervised Automatic Bias Detection for Visual Classifiers (CVPR 2025 · similarity 0.87)
- Understanding Bias in Large-Scale Visual Datasets (NeurIPS 2024 · similarity 0.86)
- Common Sense Bias Modeling for Classification Tasks (AAAI 2025 · similarity 0.86)
- Interpretable Debiasing of Vision-Language Models for Social Fairness (CVPR 2026 · similarity 0.85)
- OpenBias: Open-set Bias Detection in Text-to-Image Generative Models (CVPR 2024 · 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.
