lbourdois/OCR-liboaccn-OPUS-MIT-5M-clean

这是一个针对视觉问答任务的数据集，包含了经过处理的图像和基于不同语法形式的问题提示，图像直接以PIL对象的形式提供，而不是路径。数据集仅包含原始数据集中的法语部分，适用于视觉问答任务，特别是那些答案为OCR转录的任务。

This dataset is tailored for visual question answering tasks, including processed images and question prompts based on different grammatical forms. Images are provided directly as PIL objects rather than paths. The dataset contains only the French part of the original dataset and is suitable for visual question answering tasks, especially those with OCR transcribed answers.