In textual inversion we were used RandomHorizontalFlip transformation when getting the images in dataset. Why we didnt use again this transformation for Dreambooth?
Hi there! Good question. I’m not sure why the Dreambooth training script doesn’t use it, but I guess it could! It would be slightly less prone to overfitting, althouth the difference will be small. If you try it out for yourself, let us know how it works for you