Lucas-Y04/clevr-change-subset_1

该数据集是Lancelot53/clevr-change的一个处理子集，具有明确的count特征。每个样本包含sentences列表、sentence_count整数和配对的图像。数据集构建了一个平衡子集，每个句子计数从3到9有50个样本。特征包括sentences、sentence_count以及从CLEVR-Change继承的其他字段。预期用途包括分析多句子视觉变化理解中的推理难度，以及用于视觉语言模型的课程学习和控制评估。

This dataset is a processed subset of Lancelot53/clevr-change with an explicit count feature. Each sample contains sentences list, sentence_count integer, and paired images. A balanced subset is constructed with 50 samples for each sentence count from 3 to 9. Features include sentences, sentence_count, and other fields inherited from CLEVR-Change. Intended use includes analyzing reasoning difficulty in multi-sentence visual change understanding and curriculum learning and controlled evaluation for VLMs.