Device and method for training a model, and a vehicle

Patent metadata

Field	Value
Publication number	US-11836892-B2
Application number	US-202016992638-A
Country	US
Kind code	B2
Filing date	Aug 13, 2020
Priority date	Sep 5, 2019
Publication date	Dec 5, 2023
Grant date	Dec 5, 2023

Abstract

A device and a method for training a model including a first sub-model and a second sub-model. Digital data are down-scaled to generate first input data. The digital data are divided into multiple data areas to generate second input data. A first sub-model generates first sub-model data relating to first input data fed to it. The first sub-model data are up-scaled to form first output data. A second sub-model for the data areas generates corresponding output data areas relating to second input data fed to it. The output data areas are assembled to form second output data. The first and second output data are combined to form third output data. The first sub-model is trained on the digital data by comparing provided target data and the first output data. The second sub-model is trained on the digital data by comparing the target data and the third output data.

First claim

What is claimed is:

1. A method for the computer-implemented training of a model, which includes a first sub-model and a second sub-model, the method comprising the following steps:
downscaling digital data to generate first input data;
dividing the digital data into multiple data areas to generate second input data;
generating, by the first sub-model, first sub-model data relating to the first input data fed to the first sub-model;
up-scaling the first sub-model data to form first output data;
generating, by the second sub-model for the multiple data areas, corresponding output data areas relating to the second input data fed to the second sub-model;
assembling the output data areas to form second output data;
combining the first output data and the second output data to form third output data;
training the first sub-model by comparing provided target data, which are assigned to the digital data, and the first output data; and
training the second sub-model by comparing the target data and the third output data, wherein the model is a generative model, the digital data being associated with a first domain and the third output data being associated with a second domain, the method further comprising the following steps:
generating training data via the trained model;
training a second model using the generated training data;
processing second digital data, which are associated with the second domain, via the second model in a driving assistance system; and
determining a controlling command for an actuator of a vehicle based on the processed second digital data.

2. The method as recited in claim 1 , wherein the first sub-model data generated by the first sub-model has the same resolution as the first input data.

3. The method as recited in claim 1 , wherein the first sub-model includes a first auto-encoder.

4. The method as recited in claim 1 , wherein the second sub-model generates the second sub-model data relating to the fed second input data, for each data area of the multiple data areas an associated output data area of the multiple output data areas being generated.

5. The method as recited in claim 1 , wherein the second sub-model includes a second auto-encoder.

6. A training device configured to train a model, which includes a first sub-model and a second sub-model, the training device configured to:
downscale digital data to generate first input data;
divide the digital data into multiple data areas to generate second input data;
generate, by the first sub-model, first sub-model data relating to the first input data fed to the first sub-model;
up-scale the first sub-model data to form first output data;
generate, by the second sub-model for the multiple data areas, corresponding output data areas relating to the second input data fed to the second sub-model;
assemble the output data areas to form second output data;
combine the first output data and the second output data to form third output data;
train the first sub-model by comparing provided target data, which are assigned to the digital data, and the first output data; and
train the second sub-model by comparing the target data and the third output data, wherein the model is a generative model, the digital data being associated with a first domain and the third output data being associated with a second domain, the training device being further configured to:
generate training data via the trained model;
train a second model using the generated training data;
process second digital data, which are associated with the second domain, via the second model in a driving assistance system; and
determine a controlling command for an actuator of a vehicle based on the processed second digital data.

Assignees

• Bosch Gmbh Robert

Inventors

• Junginger Andrej

Classifications

• G06N3/094

  Adversarial learning · CPC title

• G06N3/0455

  Auto-encoder networks; Encoder-decoder networks · CPC title

• G06N3/0475

  Generative networks · CPC title

• G06N3/0495

  Quantised networks; Sparse networks; Compressed networks · CPC title

• G06N3/09

  Supervised learning · CPC title