Diffusion Models for Image Segmentation

Image segmentation is an important research area that has, due to deep learning, seen great advances in recent years. There are still problems to solve, especially when annotated data is scarce. We propose a PhD project aiming to unify agnostic segmentation models with the diffusion process. We argue this is a good idea since many of the ideas in diffusion can be applied to segmentation.

Recent diffusion model developments have been focused largely on the text-to-image domain. Adapting these methods to segmentation can give rise to useful models with human-in-the-loop or few-shot capabilities. The PhD has the potential to be valuable for collaborators of the Visual Computing section at DTU, while also having the potential for larger impacts in the research area as a whole. The applicant, Jakob Lønborg Christensen, is an honours programme student at DTU with multiple peer-reviewed publications. This PhD project would benefit significantly from not being bound to a specific application area or a specific dataset.