VALUE (Vision And Logical Understanding Evaluation) Dataset

VALUE数据集是由印度统计研究所加尔各答分所创建，包含超过200,000张标注图像，基于国际象棋游戏设计，旨在评估计算机视觉系统的逻辑理解能力。数据集中的图像展示了正在进行中的象棋游戏，每张图像都附有一套精心设计的逻辑约束规则，用于测试模型对位置和计数等关键语义能力的理解。此外，数据集还引入了额外的性能指标，以衡量模型在逻辑一致性方面的表现。该数据集适用于研究深度学习系统在遵守领域特定规则方面的能力，特别是在需要高逻辑准确性的应用领域，如机器人技术和自动化诊断等。

The VALUE Dataset was developed by the Kolkata branch of the Indian Statistical Institute. It contains over 200,000 annotated images and is designed based on the game of chess, aiming to evaluate the logical comprehension capabilities of computer vision systems. The images in the dataset depict ongoing chess games, and each image is paired with a set of carefully designed logical constraint rules, which are used to test the model's understanding of key semantic abilities including spatial positioning and counting. Furthermore, the dataset introduces additional performance metrics to measure the model's performance in terms of logical consistency. This dataset is applicable to research into the ability of deep learning systems to adhere to domain-specific rules, particularly in high-logical-accuracy application domains such as robotics and automated diagnostics.