Research map / Vision-language and multimodal / Visual question answering
Roses Are Red, Violets Are Blue... but Should VQA Expect Them To?
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 |
|---|---|---|---|
| Train BUTD on balanced instead of raw unbalanced GQA | BUTD · GQA-OOD val | Acc-tail (α=1.2): 41.9 -> 45.4 | Helped |
| Train BUTD on balanced instead of raw unbalanced GQA | BUTD · GQA-OOD val | Tot. Acc.: 59.8 -> 60.7 | No clear effect |
| Add RUBi bias-reduction training and omit its question-only branch at test time | BUTD · GQA-OOD testdev | acc-tail: 42.1 -> 35.7 | Hurt |
| Add LM bias-reduction training | BUTD · GQA-OOD testdev | acc-tail: 42.1 -> 32.2 | Hurt |
| Add BP bias-reduction training | BUTD · GQA-OOD testdev | acc-tail: 42.1 -> 30.8 | Hurt |
| Keep the RUBi question-only branch during evaluation (ablation) | BUTD + RUBi · GQA-OOD testdev | acc-tail: 35.7 -> 42.1 | Helped |
| Keep the RUBi question-only branch during evaluation (ablation) | BUTD + RUBi · GQA-OOD testdev | acc-head: 40.8 -> 49.4 | Helped |
| Add the entropy penalty to LM training (ablation) | BUTD + LM · GQA-OOD | acc-tail | Hurt |
| Use the GQA-OOD tail-accuracy metric rather than only overall accuracy to assess bias-sensitive performance | VQA evaluation · GQA-OOD val | Acc-tail: 45.4 -> 41.9 | Helped |
Closest papers by meaning
- Debiased Visual Question Answering from Feature and Sample Perspectives (NeurIPS 2021 · similarity 0.95)
- Making the v in VQA Matter: Elevating the Role of Image Understanding in Visual Question Answering (CVPR 2017 · similarity 0.94)
- Don't Just Assume; Look and Answer: Overcoming Priors for Visual Question Answering (CVPR 2018 · similarity 0.94)
- Answer Them All! Toward Universal Visual Question Answering Models (CVPR 2019 · similarity 0.93)
- When Open-Vocabulary Visual Question Answering Meets Causal Adapter: Benchmark and Approach (AAAI 2025 · similarity 0.92)
- Improving Automatic VQA Evaluation Using Large Language Models (AAAI 2024 · similarity 0.92)
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.