Outlier detection by ensembling uncertainty with negative objectness

arXiv:2402.15374v1 Announce Type: cross
Abstract: Outlier detection is an essential capability in safety-critical applications of supervised visual recognition. Most of the existing methods deliver best results by encouraging standard closed-set models to produce low-confidence predictions in negative training data. However, that approach conflates prediction uncertainty with recognition of the negative class. We therefore reconsider direct prediction of K+1 logits that correspond to K groundtruth classes and one outlier class. This setup allows us to formulate a novel anomaly score as …

abstract applications arxiv capability class confidence cs.cv cs.lg data detection low negative outlier prediction predictions recognition results safety safety-critical set standard training training data type uncertainty visual