Thanks for the article! Interesting approach!

Just a remark regarding the CNN approach. You mentioned the first idea where you just stack all the views together in the channels direction. You said you disregarded it because of the order that had to be same for all cases but you didn’t have any information on which view is which. Couldn’t you just ignored the order and you mix the orders of the views? It could even be a possible data augmentation technique since for one case (one car plug) you can have many data points (8! To be specific).