Journal of Big Data

Table 4 Results of different base-encoders using the NNCLR framework

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

Encoders	Accuracy (%)	F1-score	Parameters
Simple ConvNet	95.56	0.96	5.85 M
Skip-ConvNet	97.04	0.97	0.92 M
ShuffleNet	95.92	0.96	0.40 M

Back to article page