Research map / Vision-language and multimodal / Visual question answering
Generative Bias for Robust Visual Question Answering
Read the paperCheck 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 |
|---|---|---|---|
| Use GenB instead of the strongest listed ensemble baseline | UpDn · VQA-CP2 test | All: 57.32 -> 59.15 | Helped |
| Use GenB instead of the strongest available baseline | UpDn · VQA-CP1 test | All: 57.39 -> 62.74 | Helped |
| Add GenB debiasing | UpDn · GQA-OOD Test | All: 46.87 -> 49.43 | Helped |
| Use GenB rather than the strongest listed baseline on counterexamples | UpDn · VQA-CE | Counter: 34.41 -> 34.80 | Helped |
| Add GenB debiasing | LXMERT · VQA-CP2 test | All: 46.23 -> 71.16 | Helped |
| Add both discriminator and distillation losses to BCE-trained GenB (ablation) | GenB bias model · VQA-CP2 test | All: 56.98 -> 59.15 | Helped |
| Add discriminator loss to BCE + Distill training (ablation) | GenB bias model · VQA-CP2 test | All: 57.06 -> 59.15 | Helped |
| Add distillation loss to BCE + discriminator training (ablation) | GenB bias model · VQA-CP2 test | All: 56.54 -> 59.15 | Helped |
| Replace modified debiasing loss with GGE loss (ablation) | GenB · VQA-CP2 test | All: 59.15 -> 49.51 | Hurt |
| Replace modified debiasing loss with RUBi loss (ablation) | GenB · VQA-CP2 test | All: 59.15 -> 30.77 | Hurt |
| Replace GenB with an UpDn bias model (ablation) | UpDn target with modified loss · VQA-CP2 test | All: 52.47 -> 59.15 | Hurt |
| Replace GenB with Question-Answer bias model (ablation) | UpDn target with modified loss · VQA-CP2 test | All: 56.68 -> 59.15 | Hurt |
Closest papers by meaning
- Debiased Visual Question Answering from Feature and Sample Perspectives (NeurIPS 2021 · similarity 0.92)
- Language‑Bias‑Resilient Visual Question Answering via Adaptive Multi‑Margin Collaborative Debiasing (NeurIPS 2025 · similarity 0.91)
- Roses Are Red, Violets Are Blue... but Should VQA Expect Them To? (CVPR 2021 · similarity 0.91)
- NaturalBench: Evaluating Vision-Language Models on Natural Adversarial Samples (NeurIPS 2024 · similarity 0.88)
- FRAMES-VQA: Benchmarking Fine-Tuning Robustness across Multi-Modal Shifts in Visual Question Answering (CVPR 2025 · similarity 0.88)
- VQAttack: Transferable Adversarial Attacks on Visual Question Answering via Pre-trained Models (AAAI 2024 · similarity 0.88)
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.