The author proposes a Neural Activation Prior (NAP) for out-of-distribution detection, based on the observation that in-distribution samples induce stronger activation responses than out-of-distribution samples. This novel scoring function is simple, easy to integrate, and does not compromise classification performance.
The author proposes a Neural Activation Prior (NAP) for out-of-distribution detection, highlighting the importance of within-channel distribution information. This approach enhances OOD detection without compromising classification capabilities.