Research map / Vision-language and multimodal / Visual question answering
Yin and Yang: Balancing and Answering Binary Visual Questions
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 |
|---|---|---|---|
| Attend to question-relevant object regions instead of using holistic scene features | Q+Tuple+A-IMG vs. Q+Tuple+H-IMG · Balanced test set; balanced training | VQA accuracy: 71.03 -> 74.65 | Helped |
| Attend to question-relevant object regions instead of using holistic scene features | Q+Tuple+A-IMG vs. Q+Tuple+H-IMG · Unbalanced test set; balanced training | VQA accuracy: 69.19 -> 72.80 | Helped |
| Attend to question-relevant object regions instead of using holistic scene features | Q+Tuple+A-IMG vs. Q+Tuple+H-IMG · Complementary scene pairs; balanced training | Pair accuracy: 23.13 -> 34.73 | Helped |
| Train on balanced examples with complementary scenes | Q+Tuple+A-IMG · Balanced test set | VQA accuracy: 68.08 -> 74.65 | Helped |
| Train on balanced examples with complementary scenes | Q+Tuple+H-IMG · Balanced test set | VQA accuracy: 65.89 -> 71.03 | Helped |
| Train on balanced examples with complementary scenes | Q+Tuple+A-IMG · Unbalanced test set | VQA accuracy: 79.20 -> 72.80 | Hurt |
| Use visual input rather than language-only features | Models trained on balanced data · Balanced test set | VQA accuracy: 63.33 -> 74.65 | Helped |
| Combine question LSTM and tuple word2vec features instead of using question features alone (ablation) | Blind language-only model · Unbalanced test set; unbalanced training | VQA accuracy: 77.87 -> 78.9% | Helped |
| Combine question LSTM and tuple word2vec features instead of using tuple features alone (ablation) | Blind language-only model · Unbalanced test set; unbalanced training | VQA accuracy: 77.54 -> 78.9% | Helped |
| Provide perfect tuple extraction and object alignment (ablation) | Q+Tuple+A-IMG · Unbalanced test set | VQA accuracy: 79.2% -> 81.06% | Helped |
Closest papers by meaning
- Making the v in VQA Matter: Elevating the Role of Image Understanding in Visual Question Answering (CVPR 2017 · similarity 0.94)
- When Open-Vocabulary Visual Question Answering Meets Causal Adapter: Benchmark and Approach (AAAI 2025 · similarity 0.91)
- NaturalBench: Evaluating Vision-Language Models on Natural Adversarial Samples (NeurIPS 2024 · similarity 0.90)
- VTQA: Visual Text Question Answering via Entity Alignment and Cross-Media Reasoning (CVPR 2024 · similarity 0.90)
- Rethinking Token Reduction for Large Vision-Language Models (CVPR 2026 · similarity 0.90)
- Visual CoT: Advancing Multi-Modal Language Models with a Comprehensive Dataset and Benchmark for Chain-of-Thought Reasoning (NeurIPS 2024 · similarity 0.90)
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.