BlindTest
收藏arXiv2025-09-30 收录
下载链接：
https://vlmsareblind.github.io
下载链接
链接失效反馈官方服务：
资源简介：
该数据集包含7个非常简单的任务，旨在评估视觉语言模型（VLMs）在低级别视觉任务上的表现。该数据集对VLMs在需要精确空间信息和识别几何原语的任务上进行评估。其规模涉及多个提示和不同大小的图像。任务包括在图像中识别几何关系、计数以及识别字母。
This dataset comprises seven extremely simple tasks, designed to evaluate the performance of Vision-Language Models (VLMs) on low-level visual tasks. It assesses VLMs on tasks requiring precise spatial information and geometric primitive recognition. Its scope covers multiple prompts and images of varying sizes. The tasks include identifying geometric relationships within images, counting, and letter recognition.
