Out-of-Distribution Detection with Relative Angles

Deep learning systems deployed in real-world applications often encounter data that is different from their in-distribution (ID).