EmbSpatial-Bench
收藏arXiv2025-09-30 收录
下载链接：
https://arxiv.org/pdf/2406.05756
下载链接
链接失效反馈官方服务：
资源简介：
该数据集是一个用于评估大型视觉-语言模型（LVLMs）在具身空间理解方面的基准，它源自于覆盖了从第一人称视角的具身场景中提取的6种空间关系。该任务旨在评估在具身体验环境中对空间理解的能力。
This dataset is a benchmark for evaluating Large Vision-Language Models (LVLMs) on embodied spatial understanding. It consists of six types of spatial relations extracted from embodied scenes captured from a first-person perspective. This task is designed to assess the capability of spatial comprehension in embodied experience environments.
