How would you choose the right loss function for a computer vision problem in manufacturing defect detection?
