Table 1 Summary of model accuracy results on evaluation of consumer images where the correct image is the first result (Top-1), or within the first five results (Top-5)

From: Fast and accurate medication identification

 

Best CV avg.

Evaluation

Model

Top-1

Top-5

Top-1

Top-5

Avg precision

Parameters [millions]

Depth

Accuracy, %

Accuracy, %

All classes, %

ResNet5045

71.72

92.85

77.00

95.3

85.87

26.35

168

MobileNet53

71.7

92.00

77.10

94.40

84.29

5.04

88

SqueezeNet52

49.05

76.80

56.20

83.20

61.33

2.06

18

InceptionV354

74.42

93.30

76.30

94.80

85.94

26.56

159

  1. The table also includes average precision of the model over all classes along with its total number of parameters to assess its complexity