47 datascEYEnce! Computer Vision News When asking her for her personal lessons learned, Wenke spoke mostly about the labels having a strong inter- and intra-reader variability. It’s a problem we all have experienced before, and in Wenke’s case, it is even more problematic, as the feature to be annotated is so small and unclear. She needed to ensure that the network was as robust and consistent as possible. To achieve this, she focused on creating a consistent and meaningful ground truth and applied common techniques such as augmentation and regularisation to prevent overfitting. I wish Wenke all the best for the future and hope to see her again with some more work in deep learning for ophthalmology at the next conference!
RkJQdWJsaXNoZXIy NTc3NzU=