Published April 3, 2023 | Version v1
Conference paper Open

Runtime Monitoring for Out-of-Distribution Detection in Object Detection Neural Networks

  • 1. ROR icon Audi (Germany)
  • 2. ROR icon Technical University of Munich
  • 3. ROR icon Helmholtz Center for Information Security


Runtime monitoring provides a more realistic and applicable alternative to verification in the setting of real neural networks used in industry. It is particularly useful for detecting out-of-distribution (OOD) inputs, for which the network was not trained and can yield erroneous results. We extend a runtimemonitoring approach previously proposed for classification networks to perception systems capable of identification and localization of multiple objects. Furthermore, we analyze its adequacy experimentally on different kinds of OOD settings, documenting the overall efficacy of our approach.



Files (26.1 MB)

Name Size Download all
26.1 MB Preview Download

Additional details


Hi-Drive – Addressing challenges toward the deployment of higher automation 101006664
European Commission