Deep Open-Set Segmentation in Visual Learning

Collecting samples that exhaust all possible classes for real-world tasks is usually hard or even impossible due to many different factors. In a realistic/feasible scenario, methods should be aware that the training data is incomplete and not all knowledge is available. In this scenario, in test time, developed methods should be able to identify the unknown samples while correctly executing the proposed task to the known classes. Open-Set Recognition and Semantic Segmentation models emerge to handle this sort of scenario for visual recognition and dense labeling tasks, respectively. In this work, we propose a novel taxonomy aiming to organize the literature and provide an understanding of the theoretical trends that guided the existing approaches which may influence future methods. Moreover, we also provide the first systematic review of open-set semantic segmentation methods.