Uncertainty estimation for semantic segmentation and its application to insurance industry

01.02.2024|11:43 Uhr

It is always a great pleasure and true satisfaction for a scientist to see that a method you developed is used by others in practice, advancing an application, a business or simply people. Our industry partner ControlExpert (CE), part of Allianz SE, demonstrates how methods for uncertainty quantification in semantic segmentation can be of use in automated motor claim handling in the insurance industry.

See their recent preprint https://arxiv.org/pdf/2401.09245.pdf.

Colleagues at CE have implemented a meta classification method for semantic segmentation, which was developed in our team (https://ieeexplore.ieee.org/document/9206659) in collaboration with Volkswagen in the context of automated driving, into CE‘s products for estimating the repair cost of a given car damage. The uncertainty estimation enables more accurate predictions of car damages, discarding false positives and thus helping to solve downstream tasks such as repair cost estimation more accurately, see also https://www.autohaus.de/nachrichten/werkstatt/bilderkennung-ki-soll-schadenregulierung-beschleunigen-2713066.