Method and device for storing environmental images, in particular for training or validating data-based functions

Abstract

The invention relates to a computer-implemented method for storing environmental image data with a plurality of provided real environmental images, comprising the following steps: - Generating (S1) datasets, each of which assigns a scenario specification to an artificially generated segmented environmental image, wherein the scenario specification corresponds to a description language with which an artificially generated segmented environmental image can be created using a scenario creation tool, wherein the datasets are provided by varying the scenario specification; - Training (S2) a generative data-based scenario model with the datasets;- Determine (S3) an associated scenario specification for each of the provided real-world environment images by inverse search, in particular using a GAN inversion, in the scenario model, such that the similarity of the artificial environment image modeled for the scenario specification to the corresponding real-world environment image is maximized; - Store (S5) the determined scenario specifications and model parameters of the trained data-based scenario model as the environment image data.;

Images