Department of Image Processing

New advances in the U-net architecture

The U-net segmentation architecture has been first described in the seminal
paper of Ronneberger et al. in 2015. Since then, it has become the architecture
of choice for segmentation tasks in medical imaging and bioimaging.

With the popularity of this architecture, there have been many interesting
advancements, such as the denoising U-nets, U-net for sparse
annotations or, recently, U-net with bidirectional skips.

In the talk, Jan will try to touch some of these new versions.