Research map / Vision-language and multimodal / Visual question answering
ProMSA:Progressive Multimodal Search Agents for Knowledge-Based Visual Question Answering
ZhengXian Wu, Hangrui Xu, Kai Shi, Zhuohong Chen, Yunyao Yu, Chuanrui Zhang, ZIRUI LIAO, JunYang JunYang, Zhenyu Yang, Haonan Lu, Haoqian Wang
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 rejection-sampling SFT cold start (ablation) | ProMSA (Qwen3-VL-8B) · E-VQA (All) | accuracy: 32.8 -> 38.6 | Improved |
| add RL training with TN-GSPO after cold start (ablation) | ProMSA (Qwen3-VL-8B) · E-VQA (All) | accuracy: 38.6 -> 52.6 | Improved |
| add RL training with TN-GSPO after cold start (ablation) | ProMSA (Qwen3-VL-8B) · InfoSeek (All) | accuracy: 42.1 -> 53.4 | Improved |
| add asymmetric clipping to GRPO (ablation) | GRPO · E-VQA (All) | accuracy: 44.2 -> 49.7 | Improved |
| add asymmetric clipping to GRPO (ablation) | GRPO · InfoSeek (All) | accuracy: 43.7 -> 50.2 | Improved |
| add tool-horizon normalization (TN-GSPO) to GSPO* (ablation) | GSPO* · E-VQA (All) | accuracy: 49.3 -> 52.6 | Improved |
| add tool-horizon normalization (TN-GSPO) to GSPO* (ablation) | GSPO* · InfoSeek (All) | accuracy: 49.6 -> 53.4 | Improved |
| add image search to text-search-only agent (ablation) | ProMSA · E-VQA (All) | accuracy: 27.6 -> 52.6 | Improved |
| add text search to image-search-only agent (ablation) | ProMSA · InfoSeek (All) | accuracy: 21.4 -> 53.4 | Improved |
| add image search to text-search-only agent (ablation) | ProMSA · InfoSeek (All) | accuracy: 36.8 -> 53.4 | Improved |
| raise tool-call budget from 2/2 to 3/3 (ablation) | ProMSA · E-VQA (All) | accuracy: 48.2 -> 52.6 | Improved |
| raise tool-call budget from 3/3 to 4/4 (ablation) | ProMSA · E-VQA (All) | accuracy: 52.6 -> 52.4 | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most shared results.
- REAL — REAL: Resolving Knowledge Conflicts in Knowledge-Intensive Visual Question Answering via Reasoning-Pivot Alignment (ICML 2026) · better on 2 of 2 shared results
- DeepEyesV2 — DeepEyesV2: Toward Agentic Multimodal Model (ICLR 2026) · better on 2 of 2 shared results
- CC-VQA — CC-VQA: Conflict- and Correlation-Aware Method for Mitigating Knowledge Conflict in Knowledge-Based Visual Question Answering (CVPR 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
- Separation of Powers: On Segregating Knowledge from Observation in LLM-enabled Knowledge-based Visual Question Answering (CVPR 2025 · similarity 0.90)
- Knowledge-based Visual Question Answer with Multimodal Processing, Retrieval and Filtering (NeurIPS 2025 · similarity 0.90)
- Question Aware Vision Transformer for Multimodal Reasoning (CVPR 2024 · similarity 0.90)
- Augmenting Multimodal LLMs with Self-Reflective Tokens for Knowledge-based Visual Question Answering (CVPR 2025 · similarity 0.90)
- Visual Haystacks: A Vision-Centric Needle-In-A-Haystack Benchmark (ICLR 2025 · similarity 0.90)
- Re-ranking Reasoning Context with Tree Search Makes Large Vision-Language Models Stronger (ICML 2025 · similarity 0.89)
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.
