arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2512.06759cs.CVcs.AI
收藏

VisChainBench: 一个多轮多图视觉推理基准,超越语言先验

VisChainBench: A Benchmark for Multi-Turn, Multi-Image Visual Reasoning Beyond Language Priors

Wenbo Lyu, Yingjun Du, Jinglin Zhao, Xianton Zhen, Ling Shao

AI总结:

VisChainBench是一个多轮多图视觉推理基准,通过多代理生成流水线构建,旨在评估LVLM在连续、相互依赖任务中进行多步骤视觉推理的能力。

摘要:

理解多图、多轮场景是大型视觉-语言模型(LVLM)的关键但尚不充分的能力。现有基准主要集中在静态或横向比较上——例如发现视觉差异或评估适当性——而依赖大量语言提示。此类设置忽略了渐进、依赖上下文的推理和视觉到视觉推理的挑战。为弥补这一差距,我们提出了VisChainBench,一个大规模基准,旨在严格评估LVLM在连续、相互依赖任务中进行多步骤视觉推理的能力,且在最小语言指导下。VisChainBench包含1,457个任务,覆盖超过20,000张图像,跨越三个多样领域(例如日常场景、工程故障排除),结构上模仿真实世界决策过程。独特的是,该基准采用多代理生成流水线构建,确保高视觉多样性及受控的语言偏差。所有基准数据和构建代码均可通过以下链接查看和下载:https://huggingface.co/datasets/eyehole/VisChainBench

英文摘要:

Understanding multi-image, multi-turn scenarios is a critical yet underexplored capability for Large Vision-Language Models (LVLMs). Existing benchmarks predominantly focus on static or horizontal comparisons -- e.g., spotting visual differences or assessing appropriateness -- while relying heavily on language cues. Such settings overlook progressive, context-dependent reasoning and the challenge of visual-to-visual inference. To bridge this gap, we present VisChainBench, a large-scale benchmark designed to rigorously evaluate LVLMs' ability to perform multi-step visual reasoning across sequential, interdependent tasks with minimal language guidance. VisChainBench contains 1,457 tasks spanning over 20,000 images across three diverse domains (e.g., daily scenarios, engineering troubleshooting), structured to mimic real-world decision-making processes. Uniquely, the benchmark is constructed using a multi-agent generation pipeline, ensuring high visual diversity and controlled language bias. All the benchmark data and code for benchmark construction are available for viewing and download via following Link: https://huggingface.co/datasets/eyehole/VisChainBench

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑