Benchmarking eines Realitätsabstands zwischen Simulierten und Realen Bildgebungsdatensätzen

Details

Veröffentlichungs-Nr.

EP4718386

Anmeldetag

30. September 2024

Veröffentlichung

1. April 2026

Rechtsraum

EP

Abstract

The present disclosure enables quality benchmarking a plurality of simulated imaging datasets of a driving scene of an assisted or autonomous vehicle. The method involves processing feature vectors obtained from each simulated imaging dataset to obtain embedded feature vectors, determining local complexity indicators for multiple neighborhoods of the respective embedded feature vector, and comparing an array data structure comprising local complexity values with a reference array data structure associated with real-world imaging datasets. This comparison enables the determination of a quality benchmark value indicative of a reality gap between the plurality of simulated imaging datasets and the real-world imaging datasets, thereby allowing for improved training or validation of machine-learning models to solve control tasks of assisted or autonomous vehicles.

Anmelder

Firma

Siemens Mobility GmbH

Land

🇩🇪 Deutschland