Table 1 Average three-fold cross-validation performance of models trained on different image modalities and with different pretraining/loss schemes

From: Generalizable image segmentation for microstructure characterization through integrated SEM and EBSD analysis

 

F1 Score

HD95 (pixels)

MAE in ECD (µm)

Image modality

ImageNet DICE

MicroNet DICE

MicroNet TopoDICE

ImageNet DICE

MicroNet DICE

MicroNet TopoDICE

ImageNet DICE

MicroNet DICE

MicroNet TopoDICE

BSE 10

0.56

0.58

0.57

49.4

44.5

48.3

1.49

1.42

1.75

BSE 20

0.50

0.51

0.50

55.7

55.2

53.7

1.32

1.10

1.23

SE 10

0.60

0.62

0.62

30.6

26.3

28.7

1.08

0.68

0.57

SE 20

0.52

0.52

0.53

35.9

36.1

34.9

1.86

1.15

1.29

  1. Bolded entries denote the best value for each metric within the set of image modalities. Underlined entries denote the best value for each metric within the set of pretraining/loss schemes
Back to article page