Augmenting data when training a CNN for retinal vessel segmentation: How to warp? Artigo de Conferência