Table 1 Summary of the performance of the individual models on the ZooLake dataset.

From: Ensembles of data-efficient vision transformers as a new paradigm for automated classification in ecology

Model

No. of params for each model

Accuracy mean

F1-score mean

Arithmetic ensemble (accuracy/F1-score)

Geometric ensemble (accuracy/F1-score)

Dense121

8.1M

0.965 (3)

0.86 (1)

0.976/0.916

0.977/0.917

Efficient-B2

9.2M

0.9670 (4)

0.894 (2)

0.975/0.915

0.975/0.914

Efficient-B5

30.6M

0.964 (2)

0.87 (1)

0.971/0.891

0.971/0.898

Efficient-B6

43.3M

0.965 (1)

0.880 (7)

0.971/0.904

0.972/0.906

Efficient-B7

66.0M

0.968 (1)

0.893 (4)

0.974/0.913

0.974/0.920

Mobile-V2

3.5M

0.961 (2)

0.881 (5)

0.971/0.907

0.973/0.909

Best_6_avg

–

–

–

0.978/0.924

0.977/0.923

DeiT-Base

85.8M

0.962 (3)

0.899 (2)

0.973/0.924

0.972/0.922

  1. The ensemble score on the rightmost column is obtained by averaging across either 3 or 4 different initial conditions.
  2. The Best_6_avg model is an ensemble of DenseNet121, EfficientNet-B2, EfficientNet-B5, EfficientNet-B6, EfficientNet-B7 and MobileNet (combining learners through an arithmetic mean) models22. The numbers in parentheses are the standard errors, referred to the last significant digit.
Back to article page