opendiffusionai/cc12m-xlsd

这是一个大约包含350k张真实照片的数据集，每张照片大小在2mp到4mp之间。数据集已经通过AI清理，去除了黑白图片、非真实图片和带水印的图片等。照片的比例在3:2到2:3之间，不包括16:9的比例，因为对于通用的SD unet来说这不是必要的信息。数据集还包括长篇和短篇的描述，与CC12M子集相同。

This is a dataset consisting of approximately 350k realistic photos, each ranging from 2mp to 4mp in size. The dataset has been cleaned through AI to remove black and white images, non-real images, and images with watermarks, etc. The aspect ratios of the photos range from 3:2 to 2:3, excluding 16:9 as its considered unnecessary information for a general-purpose SD unet. The dataset also includes long and short captions, similar to those in the CC12M subsets.