Research map / Vision-language and multimodal / Multimodal LLMs
Evidence-Grounded Trustworthy Multimodal Reasoning and Evaluation Benchmark in Complex Urban Scenes
Zhaoyang Wei, Bowen Jiang, Xumeng Han, Jiashu Li, Xuehui Yu, Yuling Liu, Guorong Li, Zhenjun Han, Jianbin Jiao
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 |
|---|---|---|---|
| train with EGVOR (reflective SFT plus RL on evidence atoms) | Qwen2.5-VL-7B · AD2-Bench | average score: 55.0 -> 67.7 | Improved |
| train with EGVOR | Qwen2.5-VL-7B · AD2-Bench Basic Perception | average score: 59.3 -> 72.6 | Improved |
| train with EGVOR | Qwen2.5-VL-7B · AD2-Bench Detection | score: 51.0 -> 69.7 | Improved |
| train with EGVOR | Qwen2.5-VL-7B · AD2-Bench OCR-CER | character error rate: 15.4 -> 13.0 | Improved |
| add reflective SFT stage (ablation) | Qwen3.5-9B · AD2-Bench | average score: 59.0 -> 64.6 | Improved |
| add RL with spatial and semantic rewards after SFT (ablation) | Qwen3.5-9B · AD2-Bench | average score: 64.6 -> 69.4 | Improved |
| add reflective SFT stage (ablation) | Qwen3.5-9B · TreeBench OCR | accuracy: 75.0 -> 66.2 | Hurt |
| add reflective SFT stage (ablation) | Qwen3.5-9B · TreeBench Contact & Occlusion | accuracy: 53.7 -> 41.5 | Hurt |
| add reflective SFT stage (ablation) | Qwen3.5-9B · TreeBench Object Retrieval | accuracy: 68.8 -> 87.5 | Improved |
| add RL after SFT (ablation) | Qwen3.5-9B · TreeBench | overall accuracy: 46.5 -> 54.1 | Improved |
| add RL after SFT (ablation) | Qwen3.5-9B · TreeBench Perspective Transform | accuracy: 11.8 -> 23.5 | Improved |
| add RL after SFT (ablation) | Qwen3.5-9B · TreeBench Ordering | accuracy: 21.1 -> 38.6 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- DeepEyes-7B — DeepEyes: Incentivizing "Thinking with Images" via Reinforcement Learning (ICLR 2026) · better on 5 of 5 shared results
- Pixel-Reasoner-7B — Pixel Reasoner: Incentivizing Pixel Space Reasoning via Curiosity-Driven Reinforcement Learning (NeurIPS 2025) · better on 5 of 5 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
- Look on Demand: A Cognitive Scheduling Framework for Visual Evidence Acquisition in Multimodal Reasoning (ICML 2026 · similarity 0.90)
- GThinker: Towards General Multimodal Reasoning via Cue-Guided Rethinking (CVPR 2026 · similarity 0.89)
- See It, Say It, Sorted: An Iterative Training-Free Framework for Visually-Grounded Multimodal Reasoning in LVLMs (CVPR 2026 · similarity 0.89)
- DeepScan: A Training-Free Framework for Visually Grounded Reasoning in Large Vision-Language Models (CVPR 2026 · similarity 0.89)
- LENS: Multi-level Evaluation of Multimodal Reasoning with Large Language Models (ICLR 2026 · similarity 0.89)
- SLVR: Structured Latent Visual Reasoning via Human-like Reasoning Flows (NeurIPS 2026 · 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.
