AI Paper Atlas

Research map / Vision-language and multimodal / Multimodal LLMs

RoadBench: Benchmarking MLLMs on Fine-Grained Spatial Understanding and Reasoning under Urban Road Scenarios

Jun Zhang, Xin Zhang, Jie Feng, Long Chen, Junhui Wang, Zhicheng Liu, Depeng Jin, Yong Li

ECCV 2026 · Vision-language and multimodal › Multimodal LLMs · approach: Spatial reasoning in VLMs · 0 citations (OpenAlex)

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 changedWhereMeasuredEffect
fine-tune on RoadBench train setQwen3-VL-8B · Cross-View Lane CountingF1: 0.266 -> 0.540Improved
fine-tune on RoadBench train setQwen3-VL-8B · FPV Road Type Classificationaccuracy: 0.539 -> 0.830Improved
fine-tune on RoadBench train setQwen3-VL-8B · BEV Road Network CorrectionFrechet distance: 0.294 -> 0.154Improved
fine-tune on RoadBench train setQwen3-VL-8B · BEV Lane Designation Recognitionaccuracy: 0.372 -> 0.338Improved
compare MLLMs with domain-knowledge rule baseline on lane designationMLLMs · BEV Lane Designation Recognitionaccuracy: rule 0.587 vs GPT-5 0.579Hurt
scale parameters from GPT-5-Mini to GPT-5GPT-5 · BEV Lane CountingF1: 0.355 -> 0.305Hurt
scale Qwen2.5-VL 32B to 72BQwen2.5-VL · FPV Lane Designation Recognitionaccuracy: 0.310 -> 0.471Mixed
FPV instead of BEV viewpointMLLMs · lane counting and designationaccuracy: GPT-5 lane designation 0.579 -> 0.626Improved
cross-view instead of single-view inputMLLMs · lane countingF1: GPT-5 FPV 0.498 -> cross-view 0.400Mixed
remove image reference-line prompt (keep text only) (ablation)GLM-4.5V · BEV tasksperformanceHurt
remove text reference-line prompt (keep visual only) (ablation)GLM-4.5V · BEV tasksperformanceNo clear effect
adverse lighting in FPV images (ablation)GPT-5, Gemini-2.5-Pro, GLM-4.5V · FPV lane tasksperformanceHurt

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy