Abstract

Many recent delineation techniques owe much of theirincreased effectiveness to path classification algorithmsthat make it possible to distinguish promising paths fromothers. The downside of this development is that they re-quire annotated training data, which is tedious to produce. In this paper, we propose an Active Learning approachthat considerably speeds up the annotation process. Unlikestandard ones, it takes advantage of the specificities of thedelineation problem. It operates on a graph and can reduce the training set size by up to 80% without compromising thereconstruction quality. We will show that our approach outperforms conventional ones on various biomedical and natural image datasets, thus showing that it is broadly applicable.