Choosing an appropriate loss function is important as it affects the ability of the algorithm to produce optimum results as fast as possible.
The most important thing that the computer vision expert does is to decide which annotation type is needed to build the most accurate model(s).
State-of-the-Art Semantic Segmentation models need to be tuned in terms of memory consumption and fps output to be used in time-sensitive applications like autonomous vehicles. Here we study models like FCN, SegNets, ENets, and ICNets.
An understanding of open data sets for urban semantic segmentation shall help one understand how to proceed while training models for self-driving cars. Explore datasets like Mapillary Vistas, Cityscapes, CamVid, KITTI and DUS.