Can you classify steel surface defects as accurately as a trained convolutional network? Study the six defect classes below, then test yourself: guess each defect and compare your call against the model's, alongside a Grad-CAM saliency map showing the regions that drove its decision.