#anomaly-detection

Question 1: Single choice

An inspection model is trained only on photos of defect-free parts. At test time it must flag parts with scratches or dents it has never seen. How can such a model decide?

It models what normal parts look like and scores how far each test image deviates from that1
It learns the defect classes from labeled defective photos, like an ordinary supervised classifier2
It clusters the training photos into one group per type of defect3
It stores every training photo and flags any test photo that is not an exact copy of one4

Timed test