SafeBuild-Bench：面向建筑安全的时序鲁棒基准与图增强数据挖掘
原标题：SafeBuild-Bench: A Temporal-Robust Construction Safety Benchmark with Graph-Enhanced Data Mining
AI 摘要
SafeBuild-Bench 是一个用于评估多模态大语言模型在建筑安全场景下表现的新基准，基于超过10万条工业图像-文本记录构建，包含3314个任务实例。该基准引入了图增强多模态选择流程 GEMS，以从冗余数据中筛选信息量大的样本。当前最佳模型在基准上的得分约为60，表明现有模型在建筑安全理解方面仍不可靠。相关基准、评估脚本和代码已开源。
正文节选
Computer Science > Computer Vision and Pattern Recognition Title:SafeBuild-Bench: A Temporal-Robust Construction Safety Benchmark with Graph-Enhanced Data Mining View PDF HTML (experimental) Abstract:Construction-safety models must handle concrete deployment risks, such as a worker standing near a scaffold edge without guardrails, rather than only recognize common objects in curated images. Yet real inspection archives are redundant, long-tailed, and collected across changing sites a