testeam/Pilot_experiment
收藏Hugging Face2025-10-31 更新2025-11-01 收录
下载链接：
https://hf-mirror.com/datasets/testeam/Pilot_experiment
下载链接
链接失效反馈官方服务：
资源简介：
LLaVA-UHD-v3初步实验使用的数据集包括ShapeGrid和ShapeGrid-Sudoku基准数据集。ShapeGrid数据集旨在评估模型的空间感知能力，包括距离、面积、位置和计数等方面的问题。ShapeGrid-Sudoku数据集则用于精确评估空间方向感知，每个图像由一个3x3的网格组成，中心有一个固定的锚点，周围是随机对象。模型需要识别目标对象相对于中心的方向。
The LLaVA-UHD-v3 pilot experiment uses the ShapeGrid and ShapeGrid-Sudoku benchmark datasets. The ShapeGrid dataset is designed to evaluate the models spatial perception ability, including aspects such as distance, area, location, and count. The ShapeGrid-Sudoku dataset is used to precisely evaluate spatial directional awareness, with each image consisting of a 3x3 grid, a fixed central anchor, and surrounding random objects. The model is tasked with identifying the direction of the target object relative to the center.
提供机构：
testeam
