RGB-Th-Bench
收藏资源简介:
RGB-Th-Bench是由Aalto University、KTH Royal Institute of Technology和Detectium Oy共同开发的高质量密集型多模态基准数据集,旨在评估视觉语言模型对RGB-热成像对的理解能力。该数据集包含58个RGB-热成像样本对和1624个问题,覆盖14个核心评估或技能维度,每个样本对针对每个技能维度有四个独立的是非问题,确保从所有可能的技能维度对样本对进行彻底测试,使该基准成为一个密集的评估设置。
RGB-Th-Bench is a high-quality dense multimodal benchmark dataset jointly developed by Aalto University, KTH Royal Institute of Technology, and Detectium Oy, aiming to evaluate the comprehension capabilities of vision-language models for RGB-thermal image pairs. This dataset includes 58 RGB-thermal image sample pairs and 1624 questions, covering 14 core evaluation or skill dimensions. Each sample pair has four independent yes-no questions for each skill dimension, ensuring thorough testing of the sample pairs across all possible skill dimensions, thereby making this benchmark a dense evaluation setup.




