MuLAn

MuLAn数据集是由华为诺亚方舟实验室开发的一个创新数据集，包含超过44,860个多层注释的RGB图像，用于支持可控文本到图像生成的研究。该数据集通过一个独特的管道处理来自COCO和LAION Aesthetics 6.5的数据，生成多层RGBA分解，为高质量图像提供实例分解和遮挡信息。MuLAn数据集旨在促进新型生成和编辑技术的发展，特别是层级解决方案，为文本到图像生成AI研究开辟新途径。

The MuLAn dataset is an innovative dataset developed by Huawei Noah's Ark Lab, containing over 44,860 multi-layer annotated RGB images to support research on controllable text-to-image generation. This dataset processes data from COCO and LAION Aesthetics 6.5 via a unique pipeline to generate multi-layer RGBA decompositions, providing instance decomposition and occlusion information for high-quality images. The MuLAn dataset aims to promote the development of novel generation and editing technologies, particularly hierarchical solutions, and open up new avenues for text-to-image generation AI research.