YunxinLi/CC12M

这是一个包含多种任务的大型统一图像生成数据集，名为X2I (anything to image)，它支持多种模态的指令、主题驱动的编辑、上下文学习、计算机视觉和文本到图像生成等任务。数据集包含了多个子数据集，如时尚虚拟试穿、图像编辑、人类行为等。此外，还介绍了MJHQ-30K基准，用于评估模型的美学质量。

This is a large-scale unified image generation dataset named X2I (anything to image) that supports various tasks such as multi-modal instruction, subject-driven editing, in-context learning, computer vision, and text-to-image generation. The dataset includes several sub-datasets like fashion virtual try-on, image editing, human actions, etc. Moreover, it introduces the MJHQ-30K benchmark for evaluating the aesthetic quality of models.