ConTextual

该数据集包含人工编写的指令，这些指令针对的是富含文本的图像，并需要根据上下文进行推理。它涵盖了从自然到数字视觉场景的多样化情景。该数据集跨越八个不同的视觉场景，包含了多种任务，并采用三阶段标注过程以确保数据质量。该数据集共有506个实例，任务重点在于上下文敏感的富含文本视觉推理。

This dataset contains manually authored instructions tailored for text-rich images that require contextual reasoning. It covers diverse scenarios ranging from natural to digital visual scenes. The dataset encompasses eight distinct visual scene categories, incorporates a variety of tasks, and employs a three-stage annotation process to ensure data quality. In total, the dataset has 506 instances, with its tasks focusing on context-sensitive text-rich visual reasoning.