ManukyanD/MMEB-train-subsampled

该数据集是一个包含文本和图像对的训练数据集，主要用于文本与图像的相关性任务。数据集中的每个样本包含一个查询文本（qry）、一个正例文本（pos_text）和一个反例文本（neg_text），以及与这些文本相对应的图像路径（qry_image_path、pos_image_path、neg_image_path）。数据集分为训练集（train），共有681,995个示例，数据集大小约为36.7TB。

This dataset is a training dataset containing text and image pairs, primarily used for tasks related to the relevance between text and images. Each sample in the dataset includes a query text (qry), a positive example text (pos_text), and a negative example text (neg_text), along with corresponding image paths (qry_image_path, pos_image_path, neg_image_path). The dataset is split into a training set (train) with a total of 681,995 examples, and the dataset size is approximately 36.7TB.