Vehicle Make and Model Recognition as an Open-Set Recognition Problem and New Class Discovery

One of the main limitations traditional neural-network-based classifiers is assumption that all query data are well represented within their training set. Unfortunately, in real-life scenarios, this often not case, and unknown class may appear during testing, which drastically weakens robustness algorithms. For type problem, open-set recognition (OSR) proposes a new approach where it assumed world knowledge algorithms incomplete, so they must be prepared to detect reject objects classes. However, goal does include detection classes hidden rejected instances, would beneficial increase model’s classification capability, even after training. This paper an OSR strategy with extension for discovery aimed at vehicle make model recognition. We use neuroevolution technique contrastive loss function design domain-specific CNN generates consistent distribution feature vectors belonging same embedded space terms cosine similarity, maintaining behavior classes, serves as guide probabilistic clustering algorithm simultaneously discover The results show presented works effectively address VMMR problem furthermore able recognize objects. focused on demonstrating its effectiveness benchmark databases domain-specific. improving accuracy; however, since real-world should have strategies deal data, has been extensively addressed and, best our knowledge, never considered from perspective, work also contributes future OSR.