puccibets/fenbench

FenBench是一个全面的基准数据集，用于评估AI模型通过四种不同的视觉推理任务理解3D渲染的国际象棋位置的能力。该数据集使用3D棋盘对计算机视觉模型进行与象棋相关的空间推理测试。数据集包含200个任务，分为四个类别，每个类别都针对不同的象棋视觉理解方面进行评估：棋子识别、特定棋子类型的方块位置、棋盘上所有棋子的计数和分类、以及生成完整的棋盘位置的FEN表示。

FenBench is a comprehensive benchmark for evaluating AI models ability to understand 3D-rendered chess positions through four distinct visual reasoning tasks. The dataset uses 3D chessboards to test computer vision models on chess-related spatial reasoning. It consists of 200 tasks across four categories, each designed to assess different aspects of chess visual understanding: piece identification, square location for a specific piece type, piece counting and categorization, and generating complete board positions in FEN notation.