Need guidance with my app for archery scoring

Hi! I’m working on an app that would calculate scores by pointing the camera at a target face

I have 102 images labeled like this: 8-10-10.jpg (scores from top to bottom) and I was wondering what would be the best approach to train a model.

  1. Should I “cut” each image into 3 for each arrow or should it “learn” from the whole target with 3 arrows? (so total 306 images)
  2. If my model is capable of calculating a score from a single target image, is it possible to apply that model to a 3-target image?
  3. I also want the model to return the approximate arrow “coordinates” relative to the center of a target. Will I need to “annotate” each image with hit coordinates?
  4. Most of the images have results ranging from 10 to 8. Does it mean that the model won’t be able to recognize a 6 or 7?

Appreciate any help