Multimodal-C4 fewer-faces (mmc4-ff)

我们创建了两个额外的子集: mmc4-ff，它删除了检测到人脸的图像，mmc4-core，一个更严格过滤和缩小的语料库版本，作为开发人员的初始语料库。

We constructed two additional subsets: mmc4-ff, from which all images containing detected human faces have been removed, and mmc4-core, a more rigorously filtered and downsized version of the corpus intended as an initial corpus for developers.

搜集汇总

背景与挑战

背景概述

Multimodal-C4 fewer-faces (mmc4-ff) 是Multimodal C4语料库的一个子集，专门删除了检测到人脸的图像，以减少人脸相关数据，适用于隐私敏感或多模态预训练任务。该数据集由华盛顿大学、哥伦比亚大学和Yonsei University于2023年发布，包含692个文件，总大小约1001.0B，采用预训练ODC-BY 1.0许可证。