Table 1 Test set accuracy metrics for each of the model variants.

From: Deep learning corrosion detection with confidence

Model

Min. F1-Score

Max. F1-Score

Avg. F1-Score

FCN11

–

–

0.55

mode

raw

adj.

raw

adj.

raw

adj.

Variational

0.82

0.78

0.92

0.87

0.88

0.84

Monte-Carlo dropout

0.81

0.75

0.93

0.86

0.81

0.80

Ensemble

0.86

0.73

0.93

0.93

0.89

0.86

  1. Adjusted F1-Scores are only taken from after the 80th epoch of each fold when the variational binary cross entropy loss is initiated.
Back to article page