Generatives Design und Optimierung auf Basis von Benutzerspezifischer Terminologie für Designdeskriptoren

Details

Veröffentlichungs-Nr.

EP4790583

Anmeldetag

10. Februar 2025

Veröffentlichung

12. August 2026

Rechtsraum

EP

Abstract

A computer-implemented method that trains a generative model for supporting a user in generating or modifying a design of an object comprises a step of acquiring, from a database, a plurality of data sets of geometric bodies of objects, for each set at least one image associated with the set, wherein each data set is associated with a class of the object. The method further acquires textual parameters related to the objects, and encodes the acquired images by applying a vision-language model based on the textual parameters into vectors including numerical values for the acquired textual parameters. Then, the method generates a parametric representation of each object, and generates training data based on the parametric representations and the associated data sets and/or the images. Using the generated training data, the method then trains a generative model using the generated training data, and outputs the trained generative model.

Anmelder

Firma

HONDA MOTOR CO., LTD.

Land

🇯🇵 Japan