Fig. 5:

Effect of misregistered data on a deep metric derived from binary classification of patches. The training data was translated by 8 mm in the x direction. Top left: aggregated score of the deep classifier (deep metric) as a function of translation for a registered test case. Top right: aggregated deep metric from the classifier trained on patches with limited augmentation. Bottom: aggregated deep metric from the classifier trained on patches with heavy-augmentation.