SEED-Data-Edit - 指令引导的图像编辑混合数据集
收藏资源简介:
SEED-Data-Edit是腾讯发布的一个指令引导的图像编辑数据集。该数据集由三种不同类型的数据组成:自动化流水线产生的高质量编辑数据、从互联网收集的真实世界场景数据以及由Photoshop专家注释的高精度多轮编辑数据。数据集总量超过了350万个图像编辑对。该数据集广泛应用于图像编辑领域,特别是需要精确理解和执行编辑指令的复杂任务,如对象添加、风格转换等,其为训练语言引导的图像编辑模型提供了宝贵的资源。
SEED-Data-Edit is an instruction-guided image editing dataset released by Tencent. This dataset comprises three distinct types of data: high-quality editing data generated via automated pipelines, real-world scene data collected from the Internet, and high-precision multi-turn editing data annotated by Photoshop experts. The total scale of the dataset exceeds 3.5 million image editing pairs. This dataset is widely applied in the field of image editing, especially for complex tasks that require precise understanding and execution of editing instructions such as object addition and style transfer, and it serves as a valuable resource for training language-guided image editing models.




