Fig. 3From: Contrastive self-supervised representation learning framework for metal surface defect detectionSample images of SD-Saliency-900 dataset (a) inclusion, (b) patches, and (c) scratches, respectivelyBack to article page