I have a collection of 1000 (2.5k x 2.5k) images of jewellery on white background (my family business). Images are in 3 states - 1. unmodified jpg from camera, 2. retouched photo (dust and scratches, background removal), 3. plus colour correction (white balance, contrast, light, stone colour correction).
I’m thinking of trying to train a cnn to do 2. and 3. stage of image retouching, maybe all in one.
Any suggestions? Do I have enough data, which architecture to use, transfer learning, augmentation, other strategies? Has anyone done something similar? I don’t have big budget for training.
My knowledge is limited to Udemy nano class and all the fast.ai videos. I’m just starating to code DL, but I’m full stack web dev, so coding I like