Method and control device for generating training data for training a machine learning algorithm

1. A method for generating training data for training a machine learning algorithm, the method comprising the following steps:

providing first training data;

generating additional training data from at least one portion of the first training data, wherein the additional training data are generated in each case by applying, to all training data included in the at least one portion of the first training data, an augmentation function randomly selected from a set of possible augmentation functions; and

providing the first training data and the additional training data for training the machine learning algorithm;

wherein the generating of the additional training data from the at least one portion of the first training data further includes, in each case, randomly selecting, for all the training data included in the at least one portion of the first training data, a degree of intensity for the augmentation function from a set of possible degrees of intensity, and generating the additional training data by applying the corresponding, randomly selected augmentation function to the training data while using the randomly selected degree of intensity.

2. The method as recited in claim 1 , wherein all elements of the set of possible augmentation functions each have the same probability of being selected and/or all elements of the set of possible degrees of intensity each have the same probability of being selected.

3. The method as recited in claim 1 , wherein the first training data are image data.

4. The method as recited in claim 3 , wherein the augmentation functions include: solarization of image data, and/or posterization of image data, and/or a brightness change to image data, and/or shearing of image data, and/or rotation of image data, and/or a contrast change to image data and/or cutting of at least one portion of image data.

5. A method for training a machine learning algorithm, the method comprising:

providing first training data;

generating additional training data from at least one portion of the first training data, wherein the additional training data are generated in each case by applying, to all training data included in the at least one portion of the first training data, an augmentation function randomly selected from a set of possible augmentation functions;

providing the first training data and the additional training data for training the machine learning algorithm; and

training the machine learning algorithm based on the first training data and the additional training data;

wherein the generating of the additional training data from the at least one portion of the first training data further includes, in each case, randomly selecting, for all the training data included in the at least one portion of the first training data, a degree of intensity for the augmentation function from a set of possible degrees of intensity, and generating the additional training data by applying the corresponding, randomly selected augmentation function to the training data while using the randomly selected degree of intensity.

6. The method as recited in claim 5 , wherein the machine learning algorithm is an algorithm for image classification or an algorithm for object recognition.

7. A method for classifying image data, the method comprising the following steps:

providing first training data;

generating additional training data from at least one portion of the first training data, wherein the additional training data are generated in each case by applying, to all training data included in the at least one portion of the first training data, an augmentation function randomly selected from a set of possible augmentation functions;

providing the first training data and the additional training data for training a machine learning algorithm;

training the machine learning algorithm based on the first training data and the additional training data; and

classifying the image data using the trained machine learning algorithm;

wherein the generating of the additional training data from the at least one portion of the first training data further includes, in each case, randomly selecting, for all the training data included in the at least one portion of the first training data, a degree of intensity for the augmentation function from a set of possible degrees of intensity, and generating the additional training data by applying the corresponding, randomly selected augmentation function to the training data while using the randomly selected degree of intensity.

8. A control device for generating training data for training a machine learning algorithm, the control device comprising:

a first providing unit configured to provide first training data;

a generation unit configured to generate additional training data from at least one portion of the first training data, wherein the additional training data are generated by the generation unit by in each case applying, to all the training data included in the at least one portion of the first training data, an augmentation function randomly selected from a set of possible augmentation functions; and

a second providing unit configured to provide the first training data and the additional training data for training the machine learning algorithm;

wherein the generation unit is configured to generate the additional training data from the at least one portion of the first training data by furthermore in each case randomly selecting, for all the training data included in the at least one portion of the first training data, a degree of intensity for the augmentation function from a set of possible degrees of intensity, and generating additional training data by applying the randomly selected augmentation function to the training data while using the randomly selected degree of intensity.

9. The control device as recited in claim 8 , wherein all the elements of the set of possible augmentation functions each have the same probability of being selected and/or all the elements of the set of possible degrees of intensity each have the same probability of being selected.

10. The control device as recited in claim 8 , wherein the first training data are image data.

11. The control device as recited in claim 10 , wherein the augmentation functions include: solarization of image data, and/or posterization of image data, and/or a brightness change to image data, and/or shearing of image data, and/or rotation of image data, and/or a contrast change and/or cutting of at least one portion out of image data.

12. A control device for training a machine learning algorithm, wherein the control device is configured to train a machine learning algorithm based on training data generated by a first control device, the first control device including:

a first providing unit configured to provide first training data;

a generation unit configured to generate additional training data from at least one portion of the first training data, wherein the additional training data are generated by the generation unit by in each case applying, to all the training data included in the at least one portion of the first training data, an augmentation function randomly selected from a set of possible augmentation functions; and

a second providing unit configured to provide the first training data and the additional training data for training the machine learning algorithm;

wherein the generation unit is configured to generate the additional training data from the at least one portion of the first training data by furthermore in each case randomly selecting, for all the training data included in the at least one portion of the first training data, a degree of intensity for the augmentation function from a set of possible degrees of intensity, and generating additional training data by applying the randomly selected augmentation function to the training data while using the randomly selected degree of intensity.

13. The control device as recited in claim 12 , wherein the machine learning algorithm is an algorithm for image classification or an algorithm for object recognition.

14. A control device for classifying image data, the control device being configured to classify the image data using a machine learning algorithm, the machine learning algorithm having been trained by a first control device configured to train the machine learning algorithm based on training data generated by a second control device, the second control device including:

a first providing unit configured to provide first training data;

a generation unit configured to generate additional training data from at least one portion of the first training data, wherein the additional training data are generated by the generation unit by in each case applying, to all the training data included in the at least one portion of the first training data, an augmentation function randomly selected from a set of possible augmentation functions; and

a second providing unit configured to provide the first training data and the additional training data for training the machine learning algorithm;

wherein the generation unit is configured to generate the additional training data from the at least one portion of the first training data by furthermore in each case randomly selecting, for all the training data included in the at least one portion of the first training data, a degree of intensity for the augmentation function from a set of possible degrees of intensity, and generating additional training data by applying the randomly selected augmentation function to the training data while using the randomly selected degree of intensity.

15. A non-transitory machine-readable storage medium on which is stored a computer program for generating training data for training a machine learning algorithm, the computer program, when executed by a computer, causing the computer to perform the following steps:

providing first training data;

generating additional training data from at least one portion of the first training data, wherein the additional training data are generated in each case by applying, to all training data included in the at least one portion of the first training data, an augmentation function randomly selected from a set of possible augmentation functions; and

providing the first training data and the additional training data for training the machine learning algorithm;

wherein the generating of the additional training data from the at least one portion of the first training data further includes, in each case, randomly selecting, for all the training data included in the at least one portion of the first training data, a degree of intensity for the augmentation function from a set of possible degrees of intensity, and generating the additional training data by applying the corresponding, randomly selected augmentation function to the training data while using the randomly selected degree of intensity.