Research map / Vision-language and multimodal / Visual question answering
Making the v in VQA Matter: Elevating the Role of Image Understanding in Visual Question Answering
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 |
|---|---|---|---|
| Evaluate models trained on the unbalanced dataset on the balanced dataset | VQA models · VQA balanced val | accuracy: 60.36 -> 54.22 | Hurt |
| Train on the balanced dataset instead of the unbalanced dataset | VQA models · VQA balanced val | accuracy: 54.22 -> 56.08 | Helped |
| Use the full balanced training set instead of the half-size balanced training set | VQA models · VQA balanced val | accuracy: 56.08 -> 59.14 | Helped |
| Train on balanced data rather than unbalanced data | HieCoAtt · Complementary image pairs | both-pairs-correct rate: 13.5% -> 17.7% | Helped |
| Train on balanced data rather than unbalanced data | HieCoAtt · Complementary image pairs | identical prediction rate: 59.9% -> 49.4% | Helped |
| Balance answers per question using complementary images with different answers | VQA models · VQA balanced val | accuracy: 60.36 -> 54.22; 54.22 -> 56.08 | Mixed |
| Train the counter-example explanation model instead of using distance ranking | Counter-example retrieval · Counter-example retrieval | Recall@5: 42.84 -> 43.39 | Helped |
| Train the counter-example explanation model instead of random ranking | Counter-example retrieval · Counter-example retrieval | Recall@5: 20.79 -> 43.39 | Helped |
| Train the counter-example explanation model instead of VQA probability ranking | Counter-example retrieval · Counter-example retrieval | Recall@5: 21.65 -> 43.39 | Helped |
| Use a question-only model without image input | VQA models · VQA balanced val | accuracy: 59.14 -> 43.01 | Hurt |
Closest papers by meaning
- Don't Just Assume; Look and Answer: Overcoming Priors for Visual Question Answering (CVPR 2018 · similarity 0.95)
- Roses Are Red, Violets Are Blue... but Should VQA Expect Them To? (CVPR 2021 · similarity 0.94)
- Debiased Visual Question Answering from Feature and Sample Perspectives (NeurIPS 2021 · similarity 0.94)
- When Open-Vocabulary Visual Question Answering Meets Causal Adapter: Benchmark and Approach (AAAI 2025 · similarity 0.94)
- VTQA: Visual Text Question Answering via Entity Alignment and Cross-Media Reasoning (CVPR 2024 · similarity 0.93)
- CVQA: Culturally-diverse Multilingual Visual Question Answering Benchmark (NeurIPS 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.