Fig. 10From: Contrastive self-supervised representation learning framework for metal surface defect detectionValidation accuracy curve of convolutional encoder: baseline, pretraining, and downstream/finetuningBack to article page