Dataset
Could you please explain how the data was created?
Using existing ojbect removal tool to create image pairs
Do you have any demo code you can share that shows how to use this?
For flux fill, the input is the original image covered by the mask, so only the background image and the mask are needed?
Do you have any demo code you can share that shows how to use this?
Just use the workflow in comfyui or use diffusers with fill pipeline and add a lora
For flux fill, the input is the original image covered by the mask, so only the background image and the mask are needed?
Three images, Ground True which is no subject, Factual which is subject existed and the mask of the subject. (For training dataset) For more information, you could refer to this https://objectdrop.github.io/
For flux fill, the input is the original image covered by the mask, so only the background image and the mask are needed?
Three images, Ground True which is no subject, Factual which is subject existed and the mask of the subject. (For training dataset) For more information, you could refer to this https://objectdrop.github.io/
Thank you for your patient response. Objectdrop used approximately 2,500 samples. Is your training data of a similar scale? I trained on 350 pairs of data, but the results are not as expected.