UNet originally published by Ronneberger et al. is one of the most popular CNN architecture for Biomedical image segmentation. A follow up work by Zhou et al. makes an improvement to this architecture. Discussed in the following blog post here is a TL;DR:
UNet++ aims to improve segmentation accuracy, with a series of nested, dense skip pathways.
Redesigned skip pathways made optimisation easier with the semantically similar feature maps.
Dense skip connections improve segmentation accuracy and improve gradient flow.
Deep supervision allows for model complexity tuning to balance between speed and performance optimisation.