Figure 3 | Scientific Reports

Figure 3

From: A vision transformer architecture for the automated segmentation of retinal lesions in spectral domain optical coherence tomography images

Figure 3

Comparison of the methods’ computational efficiency across four different metrics. The mean cv5-Dice score over the classes is plotted against the total number of trainable parameters (top-left), the number of multiply-accumulate operations (MACs) (top-right), the measured GPU memory used during training (bottom-left), and the mean training duration over the folds (bottom-right). The dotted line represents the relationship between the two methods with the lowest and highest Dice scores.

Back to article page