Journal of Big Data

Table 2 Self-supervised pretraining and finetuning using SimCLR framework

From: Contrastive self-supervised representation learning framework for metal surface defect detection

Encoders	Accuracy (%)	F1-score	Parameters
SqueezeNet	88.89	0.88	0.13 M
ResNet50	91.11	0.91	49.80 M
ShuffleNet	97.04	0.97	0.41 M
Skip-ConvNet	97.41	0.97	0.92 M
ConvNet	97.78	0.98	6.01 M

Back to article page