Table 2 Comparison of the identification model architectures.

From: Deep learning for automated materials characterisation in core-loss electron energy loss spectroscopy

Model

# of parameters [mln.]

Simulated test set

Experimental test set

Inference time [s]

F\(_1\)-score

EMR

RMSE

F\(_1\)-score

EMR

RMSE

MLP

56

0.50

0.05

0.16

0.42

0.12

0.15

0.1

CNN

45

0.90

0.68

0.07

0.76

0.42

0.12

0.2

ResNet

41

0.89

0.68

0.07

0.77

0.53

0.10

0.3

U-Net

20

0.86

0.63

0.07

0.84

0.62

0.09

0.2

ViT

2

0.84

0.55

0.09

0.84

0.60

0.08

0.6

CCT

5

0.87

0.60

0.08

0.79

0.58

0.08

0.6