This repository contains some of the work produced as part of my research internship at Safran Tech.
The aim of this internship was to develop a method for unsupervised domain adaptation of semantic segmentation algorithms to difficult weather conditions (fog, rain, snow, night). Unsupervised domain adaptation avoids the need to annotate the target domain, in this case images in harsh weather conditions.
Recent methods have drastically reduced the gap between adapted and supervised models, and in particular Transformers have proved to be excellent candidates for adaptation.
The internship oral presentation in this repository summarises all the work carried out. Don't hesitate to contact me if you want more informations.