V-LoL

V-LoL是一个专为视觉逻辑学习设计的诊断数据集，由达姆施塔特工业大学的人工智能与机器学习研究组创建。该数据集包含11个子集，共计超过12,000个实例，用于测试和评估AI模型在视觉逻辑学习方面的能力。V-LoL通过结合复杂的视觉场景和灵活的逻辑推理任务，提供了一个多功能的框架，用于研究广泛的视觉逻辑学习挑战。数据集中的每个实例不仅包括视觉上吸引人的图像，还附带详细的场景信息，如对象属性、掩码、边界框、3D场景位置和深度信息。V-LoL特别适用于评估和比较符号AI、神经AI和神经符号AI模型，揭示不同方法的优点和局限性。

V-LoL is a diagnostic dataset specifically designed for visual logic learning, created by the Artificial Intelligence and Machine Learning Research Group at Technische Universität Darmstadt. This dataset comprises 11 subsets with over 12,000 instances in total, which is used to test and evaluate the visual logic learning capabilities of AI models. By combining complex visual scenes and flexible logical reasoning tasks, V-LoL provides a versatile framework for studying a wide range of visual logic learning challenges. Each instance in the dataset not only includes visually engaging images but also comes with detailed scene information, such as object attributes, masks, bounding boxes, 3D scene positions and depth information. V-LoL is particularly suitable for evaluating and comparing symbolic AI, neural AI and neuro-symbolic AI models, revealing the strengths and limitations of different approaches.