1 paper
Daniil Karzanov, Marcin Detyniecki
Deep neural networks for image classification often exhibit overconfidence on out-of-distribution (OOD) samples. To address this, we introduce Geometrically Constrained Outlier Syn…