TIGER-Lab/OmniEdit-Filtered-1.2M

OmniEdit数据集是一个用于处理七种不同图像编辑任务的通用编辑器。数据集包含源图像和编辑后的图像，以及编辑指令、任务类型、语义一致性评分、感知质量评分和总体评分等信息。数据集通过专家蒸馏合成，并经过VIEScore过滤，包含120万对图像编辑对。

The OmniEdit dataset is a versatile image editing dataset designed to handle seven different image editing tasks, including addition, removal, background, environment, style, etc. The dataset includes source and edited images, editing instruction lists, editing task categories, semantic consistency scores, perceptual quality scores, and overall scores, along with the reasoning for these scores. The dataset is synthesized through specialist distillation, containing 1.2M image pairs covering seven different editing skills, and has been filtered with VIEScore.