UrBench
收藏资源简介:
UrBench是由上海人工智能实验室等机构创建的综合性城市环境评估数据集,包含11.6K条问题,覆盖地理定位、场景推理、场景理解和对象理解四个维度。数据集通过结合现有数据集和收集自11个城市的数据,利用跨视图检测匹配方法创建新的标注,并整合多种方法生成高质量问题。UrBench主要用于评估和提升大型多模态模型在城市环境中的应用能力,特别是在多视角理解方面的性能。
UrBench is a comprehensive urban environment evaluation dataset created by Shanghai AI Laboratory and other institutions. It comprises 11.6K questions covering four core dimensions: geolocation, scene reasoning, scene understanding, and object understanding. The dataset is developed by integrating existing datasets and data collected from 11 cities, adopting cross-view detection and matching approaches to generate new annotations, and combining multiple methods to produce high-quality questions. UrBench is primarily designed to evaluate and improve the application capabilities of large multimodal models in urban environments, particularly their performance in multi-view understanding.




