Research map / Vision-language and multimodal / Visual question answering
REAL: Resolving Knowledge Conflicts in Knowledge-Intensive Visual Question Answering via Reasoning-Pivot Alignment
Kai Ye, Xianwei Mao, Sheng Zhou, Zirui Shao, Ye Mo, Liangliang Liu, Haikuan Huang, Bin Li, Jiajun Bu
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 RPGD contrastive decoding on top of RPA-SFT (ablation) | REAL (Qwen3-VL-8B) · InfoSeek All | accuracy: 38.7 -> 44.1 | Improved |
| add RPGD contrastive decoding on top of RPA-SFT (ablation) | REAL (Qwen3-VL-8B) · E-VQA All | accuracy: 38.1 -> 41.4 | Improved |
| add RPGD contrastive decoding on top of RPA-SFT (ablation) | REAL (InternVL3.5-8B) · InfoSeek All | accuracy: 36.2 -> 43.8 | Improved |
| add RPGD contrastive decoding on top of RPA-SFT (ablation) | REAL (LLaVA-1.5-7B) · E-VQA All | accuracy: 28.5 -> 30.9 | Improved |
| use accurate Stage 1 conflict signal instead of uniform signal (ablation) | RPGD (Qwen3-VL-8B) · InfoSeek All | accuracy: 39.0 -> 44.1 | Improved |
| use accurate Stage 1 conflict signal instead of random signal (ablation) | RPGD (Qwen3-VL-8B) · E-VQA All | accuracy: 39.3 -> 41.4 | Improved |
| add patch shuffle (ablation) | RPGD (Qwen3-VL-8B) · E-VQA All | accuracy: 39.2 -> 41.4 | Improved |
| add adaptive gating (ablation) | RPGD (Qwen3-VL-8B) · E-VQA All | accuracy: 39.5 -> 41.4 | Improved |
| add Gram-Schmidt orthogonalization (ablation) | RPGD (Qwen3-VL-8B) · E-VQA All | accuracy: 38.9 -> 41.4 | Improved |
| use patch shuffle instead of blank image as perturbation (ablation) | RPGD (Qwen3-VL-8B) · InfoSeek All | accuracy: 37.4 -> 44.1 | Improved |
| use patch shuffle instead of Gaussian blur (ablation) | RPGD (Qwen3-VL-8B) · E-VQA All | accuracy: 38.1 -> 41.4 | Improved |
| use RPGD instead of VCD decoding (ablation) | RPA-SFT (LLaVA-1.5-7B) · E-VQA All | accuracy: 26.0 -> 30.9 | Improved |
Later papers that compared with this method
1 later accepted paper reports REAL as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- ProMSA:Progressive Multimodal Search Agents for Knowledge-Based Visual Question Answering (ECCV 2026 · better on 2 of 2 shared results)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- CC-VQA: Conflict- and Correlation-Aware Method for Mitigating Knowledge Conflict in Knowledge-Based Visual Question Answering (CVPR 2026 · similarity 0.93)
- StaR-KVQA: Structured Reasoning Traces for Implicit-Knowledge Visual Question Answering (CVPR 2026 · similarity 0.91)
- Identifying and Resolving Pitfalls of Knowledge-Based VQA Benchmarks: Auditing, Repairing, and Augmenting (ECCV 2026 · similarity 0.91)
- GQA: A New Dataset for Real-World Visual Reasoning and Compositional Question Answering (CVPR 2019 · similarity 0.90)
- Separation of Powers: On Segregating Knowledge from Observation in LLM-enabled Knowledge-based Visual Question Answering (CVPR 2025 · similarity 0.89)
- When Open-Vocabulary Visual Question Answering Meets Causal Adapter: Benchmark and Approach (AAAI 2025 · 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.
