yzwang/X2I-text-to-image

X2I数据集是一个大规模的统一图像生成数据集，旨在训练具有强大多任务处理能力的OmniGen模型。该数据集首次以统一格式创建，涵盖了多模态指令、主题驱动编辑、上下文学习、计算机视觉和文本到图像生成等多个子任务。特别是，文本到图像生成部分包括了laion-coco-aesthetic子集，该子集从LAION-COCO数据集中筛选出4,134,263张图像（图像尺寸大于384x384，美学评分大于4.75，水印概率小于0.5）。此外，还包括了其他多个开源数据集，如Recap-DataComp-1B、SAM-LLaVA-Captions10M、ALLaVA-4V、DOCCI、ShareGPT4V、DenseFusion-1M和JourneyDB。

The X2I dataset is a large-scale unified image generation dataset, providing a unified format dataset for the first time, suitable for various tasks such as multi-modal instruction, subject-driven editing, in-context learning, computer vision, and text-to-image generation. The X2I-text-to-image subset is a filtered subset based on the LAION-COCO dataset, containing 4,134,263 images, and incorporates other open-source datasets.