The training dataset covers a synthetically generated ~20K editing triplets on changing color, changing object, adding an object, removing an object and changing texture.
Dataset Structure
"img_id" (str): unique id
"source_img" (str): the original image generated using FLUX
"instruction" (str): edit instruction of how the input image should be changed.
"target_img" (str): the edited image corresponding to the input image and… See the full description on the dataset page: https://huggingface.co/datasets/bpathir1/RefEdit.
