Rendered models of variations of donuts were used to train object detection network, to count them on the production line