Bionic Vision Lab @bionicvisionlab.org · Sep 27 #DeepLearning is routinely used in fundus image analysis, but doesn't work so well for the degenerated retina, where drusen, lesions, and other abnormalities are common. To address this, we enriched a U-Net backbone with hierarchical bottleneck attention (HBA) to form HBA-U-Net:
0 likes 1 replies
?
Replies
Our novel HBA block combines and refines self attention, channel attention, and relative-position attention to highlight retinal abnormalities important for landmark detection in the degenerated retina