I am trying to train a ControlNetModel that takes a conditioning_image, a target_image and a prompt, I have the images in 2 folders and a metadata.jsonl file that has a json object in each line with source, target and the prompt, I am getting an error - āfile_name must be present as a dictionary key in metadata filesā. Does anyone know how exaclty i need to structure my dataset, or any links to datasets that work.
Did you solve it? I have the same error.
We can try to generate a ājsonā dataset, use the image path instead of the image in the dataset, and then edit the training code: load the dataset from ājsonā and get the image path, and finally read the image into memory.
reference: Load