Kapitel/Beitrag aus dem Buch: Längle T. & Heizmann M. 2024. Forum Bildverarbeitung 2024.
Deep neural networks achieve outstanding results in perception tasks such as semantic segmentation and monocular depth estimation, making them indispensable in safety-critical applications like autonomous driving and industrial inspection. However, they often suffer from overconfidence and poor explainability, especially for out-of-domain data. While uncertainty quantification has emerged as a promising solution to these challenges, multi-task settings still need to be investigated in this regard. In an effort to shed light on this, we evaluate onte Carlo Dropout, Deep Sub-Ensembles, and Deep Ensembles for joint semantic segmentation and monocular depth estimation. Thereby, we reveal that Deep Ensembles stand out as the preferred choice and show the potential benefit of multi-task learning with regard to the uncertainty quality in comparison to solving both tasks separately.
Landgraf, S et al. 2024. Evaluation of multi-task uncertainties in joint semantic segmentation and monocular depth estimation. In: Längle T. & Heizmann M (eds.), Forum Bildverarbeitung 2024. Karlsruhe: KIT Scientific Publishing. DOI: https://doi.org/10.58895/ksp/1000174496-13
This chapter distributed under the terms of the Creative Commons Attribution + ShareAlike 4.0 license. Copyright is retained by the author(s)
Dieses Buch ist Peer reviewed. Informationen dazu Hier finden Sie mehr Informationen zur wissenschaftlichen Qualitätssicherung der MAP-Publikationen.
Veröffentlicht am 21. November 2024