Table 1 ML Model Results on the Test Set of 20 experimental images

From: Accelerating domain-aware electron microscopy analysis using deep learning models with synthetic data and image-wide confidence scoring

Training data, domain filtering

Selected weights (epoch number)

Unfiltered images

CV metrics

Engineering metrics (swelling)

P

R

F1

R2

RMSE (%)

Experimental

Best (700)

20

0.86

0.64

0.71

0.81

0.14

Average (400–1000)

20

0.87 ± 0.01

0.63 ± 0.02

0.70 ± 0.02

0.73 ± 0.11

0.17 ± 0.05

Synthetic

Best (700)

20

0.86

0.63

0.71

0.82

0.14

Average (400–1000)

20

0.81 ± 0.03

0.62 ± 0.02

0.69 ± 0.02

0.70 ± 0.15

0.18 ± 0.04

Experimental, with Self-Regulation

Best (700)

16

0.91

0.73

0.80

0.91

0.10

Average (400–1000)

17

0.90 ± 0.01

0.69 ± 0.02

0.76 ± 0.02

0.87 ± 0.03

0.12 ± 0.01

Synthetic, with Self-Regulation

Best (700)

15

0.91

0.72

0.80

0.91

0.10

Average (400–1000)

15

0.87 ± 0.03

0.68 ± 0.02

0.76 ± 0.02

0.84 ± 0.07

0.14 ± 0.06

  1. Values are calculated by taking the mean of all image values for a given metric. ‘Best’ represents the performance of a single weights file. ‘Average’ is the mean performance of 7 evenly spaced sets of weights between the given epochs. The uncertainty for the “Average” values is the standard deviation of the 7 values that were averaged.
Back to article page