Web-Based Images
收藏arXiv2025-09-30 收录
下载链接：
https://github.com/facebookresearch/ssl-data-curation
下载链接
链接失效反馈官方服务：
资源简介：
该数据集是从公开可用的网络爬虫数据仓库中精心挑选的一组网络图片，经过筛选以去除不安全内容和重复项。在用于自监督学习之前，这些图片经过了处理，去除了有害内容，保护了隐私，并确保了图片质量。该数据集规模达到了7.43亿张独特的图片，用于自监督学习的预训练任务。
This dataset is a meticulously curated collection of web images selected from publicly available web crawler data repositories, with initial filtering to remove unsafe content and duplicate entries. Prior to being utilized for self-supervised learning pre-training tasks, these images underwent additional processing to eliminate harmful content, safeguard privacy, and guarantee consistent image quality. Comprising 743 million unique images, this dataset is specifically employed for self-supervised learning pre-training tasks.
提供机构：
Publicly available repository of crawled web data
