TIGER-Lab/ABC-Pretraining-Data

ABC预训练数据集，用于视觉问答和图像文本到文本的任务。该数据集基于Google的概念性字幕数据集，每个数据项包括一个图像下载链接和每个项的否定样本。数据集总大小约为300GB。ABC pretraining dataset for visual question answering and image-text-to-text tasks. This dataset is derived from Googles Conceptual Captions, with each item including an image download URL and negative samples for each item. The entire dataset is approximately 300GB in size.