Given a specific vision task, how would you design the data augmentation pipeline to handle lighting variations for Qualcomm models?
