Metric learning for novelty and anomaly detection

M. Masana, I. Ruiz, J. Serrat, J. Van De Weijer, A.M. Lopez

Producció científica: Contribució a revistaArticleRecercaAvaluat per experts

26 Cites (Scopus)
1 Descàrregues (Pure)

Resum

When neural networks process images which do not resemble the distribution seen during training, so called out-of-distribution images, they often make wrong predictions, and do so too confidently. The capability to detect out-of-distribution images is therefore crucial for many real-world applications. We divide out-of-distribution detection between novelty detection -images of classes which are not in the training set but are related to those-, and anomaly detection -images with classes which are unrelated to the training set. By related we mean they contain the same type of objects, like digits in MNIST and SVHN. Most existing work has focused on anomaly detection, and has addressed this problem considering networks trained with the cross-entropy loss. Differently from them, we propose to use metric learning which does not have the drawback of the softmax layer (inherent to cross-entropy methods), which forces the network to divide its prediction power over the learned classes. We perform extensive experiments and evaluate both novelty and anomaly detection, even in a relevant application such as traffic sign recognition, obtaining comparable or better results than previous works.
Idioma originalAnglès
Pàgines (de-a)2176-2181
Nombre de pàgines6
RevistaProceedings of the British Machine Vision Conference
Estat de la publicacióPublicada - 2019

Fingerprint

Navegar pels temes de recerca de 'Metric learning for novelty and anomaly detection'. Junts formen un fingerprint únic.

Com citar-ho