Training Dataset of Image Pairs

该数据集是一个大规模的训练集，包含了近70亿张图像对，覆盖了多轮次和长上下文生成中的8种基本任务类型。数据集中包含了用于将一张图像转换成另一张图像的标注自然语言指令，这些指令是通过模板方法和基于MLLM的方法相结合创建的。该数据集的规模大约为70亿图像对，涉及的任务是视觉生成和编辑任务。

This is a large-scale training dataset containing nearly 7 billion image pairs, covering 8 basic task types in multi-turn and long-context generation. The dataset includes annotated natural language instructions for converting one image into another, which were created via a combination of template-based methods and MLLM-based approaches. The total scale of this dataset is approximately 7 billion image pairs, with its involved tasks being visual generation and editing tasks.