Object detection with extremely limted, yet accurate training data

Question

Fast rundown:

I\'m trying to train an object detection algorithm to detect a vast array of different objects (200+), with extremely limited trai