ContinuousEditControl
收藏资源简介:
ContinuousEditControl数据集是一个由110,000张图像组成的合成数据集,用于训练Kontinuous Kontext模型。该数据集包含了图像、编辑指令、编辑强度和相应的目标编辑图像。数据集的生成过程包括使用Qwen LVLM生成编辑指令,应用Flux Kontext模型生成全强度编辑图像,然后使用Freemorph模型生成不同强度的中间编辑图像。数据集的创建旨在为指令驱动图像编辑提供连续的编辑强度控制,使得用户可以平滑地调整编辑程度。
The ContinuousEditControl dataset is a synthetic dataset composed of 110,000 images, intended for training the Kontinuous Kontext model. It includes images, editing instructions, editing strengths, and their corresponding target edited images. The dataset is generated through a three-stage pipeline: first, generating editing instructions using the Qwen LVLM; second, generating full-strength edited images with the Flux Kontext model; third, generating intermediate edited images with varying editing strengths via the Freemorph model. This dataset is developed to enable continuous editing strength control for instruction-driven image editing, allowing users to smoothly adjust the degree of image editing.




