Web Image-Text Pairs
收藏arXiv2025-09-30 收录
下载链接：
https://github.com/salesforce/ALBEF/
下载链接
链接失效反馈官方服务：
资源简介：
该数据集是一个大规模的数据集，用于训练ALBEF模型，包含了大量的网络图像-文本对，这些对可能存在一定的噪声。ALBEF模型能够利用这些含有噪声的网络数据来提升预训练的性能，而且它不需要边框标注或高分辨率的图像。该数据集的规模从400万到1400万张图像不等，其任务是进行视觉-语言表征学习。
This is a large-scale dataset designed for training the ALBEF model, which comprises a large number of web-collected image-text pairs that may contain certain noise. ALBEF can utilize this noisy web data to enhance pre-training performance, and it does not require bounding box annotations or high-resolution images. The scale of this dataset ranges from 4 million to 14 million images, and its target task is visual-language representation learning.
