AIML-TUDA/CLEVR-Sudoku

CLEVR-Sudoku是一个挑战性的视觉谜题数据集，要求模型具备视觉对象感知和推理能力。每个样本包含一个部分填充的基于CLEVR样式的数独谜题图像，以及单独的选项图像，这些图像展示了对象属性如何映射到数字。同时还包括了指定谜题属性和真实解决方案的元数据。数据集共包含6000个谜题，分为6个不同的子集，用于视觉推理和模式识别任务。

CLEVR-Sudoku is a challenging visual puzzle dataset that requires both visual object perception and reasoning capabilities. Each sample includes a partially filled Sudoku puzzle presented with CLEVR-based imagery, separate options images that illustrate how specific object properties map to digits, and metadata specifying puzzle attributes and the ground-truth solution. The dataset contains a total of 6,000 puzzles across 6 different subsets for visual reasoning and pattern recognition tasks.