Fig. 9.
Hybrid DL/ML models (green) and end-to-end DL models (red) are shown in the above scatter plot based on their accuracy on both the ‘Field’ dataset (y-axis) and the ‘Lab’ dataset (x-axis). Top-performing end-to-end DL models fall on the Pareto Optimal Frontier [39]: InceptionV3 + FC, the CNN ensemble, and VGG-16 + FC.