Research map / Vision-language and multimodal / Visual question answering
Beyond Single-View Indexing: Structure-Aware Multi-View Retrieval for Knowledge-Based VQA
Hao Wang, Xujia Li, Lei Chen
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 similarity propagation (Module I) to naive multi-view averaging (ablation) | SCAR · E-VQA | Recall@20: 56.4 -> 63.8 | Improved |
| add cross-view redundancy regulation (Module II) on top of Module I (ablation) | SCAR · E-VQA | Recall@20: 63.8 -> 65.1 | Improved |
| add Module II on top of Module I (ablation) | SCAR · E-VQA | Recall@5: 48.3 -> 49.9 | Improved |
| add summary view to image view (ablation) | multi-view retrieval · E-VQA | Recall@20: 48.8 -> 62.9 | Improved |
| add title view to image+summary (ablation) | multi-view retrieval · E-VQA | Recall@20: 62.9 -> 65.1 | Improved |
| combine summary and title views (both text-based) (ablation) | multi-view retrieval · E-VQA | Recall@20: 58.7 -> 54.9 | Hurt |
| use SCAR instead of OMGM with the same reranker | SCAR · E-VQA | Recall@1: 39.0 -> 43.1 | Improved |
| use SCAR instead of OMGM with reranking | SCAR · E-VQA | Recall@20: 58.7 -> 65.1 | Improved |
| use SCAR retrieval instead of best single view for answering | LLaVA-1.5-7B · E-VQA | BEM score: 28.8 -> 34.4 | Improved |
| use SCAR retrieval instead of best single view for answering | Qwen3-VL-8B · E-VQA | BEM score: 33.4 -> 39.8 | Improved |
| use SCAR retrieval instead of best single view for answering | InternVL3.5-8B · E-VQA | BEM score: 32.4 -> 39.0 | Improved |
| use SCAR retrieval instead of summary view for answering | Qwen3-VL-8B · InfoSeek All | VQA accuracy: 32.1 -> 34.5 | Improved |
Closest papers by meaning
- Separation of Powers: On Segregating Knowledge from Observation in LLM-enabled Knowledge-based Visual Question Answering (CVPR 2025 · similarity 0.88)
- Re-ranking Reasoning Context with Tree Search Makes Large Vision-Language Models Stronger (ICML 2025 · similarity 0.88)
- When Open-Vocabulary Visual Question Answering Meets Causal Adapter: Benchmark and Approach (AAAI 2025 · similarity 0.87)
- VQA4CIR: Boosting Composed Image Retrieval with Visual Question Answering (AAAI 2025 · similarity 0.87)
- Retrieval-Augmented Visual Question Answering via Built-in Autoregressive Search Engines (AAAI 2025 · similarity 0.87)
- FRAMES-VQA: Benchmarking Fine-Tuning Robustness across Multi-Modal Shifts in Visual Question Answering (CVPR 2025 · similarity 0.87)
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.
