Determination of the decision-relevant image components for an image classifier using binary masks

1. A method for measuring components of an input image on which an image classifier bases its decision regarding an assignment of the input image to one or multiple class(es) of a predefined classification, the method comprising the following steps:

providing binary masks which indicate which pixels of the input image and/or of an intermediate product formed in the image classifier, are considered relevant;

assessing each binary mask of the binary masks using a quality function, which is a measure of an extent to which at least one classification score, supplied by the image classifier, with respect to at least one target class changes when the pixels of the input image or of the intermediate product, which are relevant according to the binary mask, are changed; and

ascertaining sought-after components of the input image relevant for the decision of the image classifier from a combination of the binary masks with their respective assessments by the quality function, wherein at least one derivative and/or one gradient of the at least one classification score with respect to the pixels of the input image or of the intermediate product which are relevant according to the binary mask has an improving effect on a value of the quality function.

2. The method as recited in claim 1 , wherein the intermediate product is selected which results from the input image due to processing in one or multiple convolutional layer(s).

3. The method as recited in claim 2 , wherein the intermediate product is selected which is mapped on the at least one classification score in the image classifier by a classifier layer.

4. The method as recited in claim 1 , wherein a dependence of the at least one classification score on the pixels of the input image or of the intermediate product, which according to the binary mask are relevant, has an improving effect on a value of the quality function.

5. The method as recited in claim 4 , wherein a dependence of the at least one classification score on the pixels of the input image or of the intermediate product which according to the binary mask are not relevant has/have a worsening effect on the value of the quality function.

6. The method as recited in claim 1 , wherein a derivative and/or a gradient of the at least one classification score with respect to the pixels of the input image or of the intermediate product which are not relevant according to the binary mask has a worsening effect on the value of the quality function.

7. The method as recited in claim 1 , wherein at least one gradient remains unconsidered in response to its absolute value being below a predefined threshold.

8. The method as recited in claim 1 , wherein at least one gradient and/or at least one contribution of a pixel to the gradient is discretized in that it, in response to being above a predefined threshold, is set to a first value and in that it, in response to being below the predefined threshold negated, is set to a second value.

9. The method as recited in claim 1 , wherein the quality function includes at least one classification score which is supplied by the image classifier when the input image or the intermediate product is changed for the pixels relevant according to the binary mask to the effect that,

in response to a contribution of a pixel to the gradient of the classification score with respect to the relevant pixels being above a predefined threshold, the pixel is changed to a first predefined substitute value, and

in response to a contribution of a pixel to the gradient of the classification score with respect to the relevant pixels being below the negated predefined threshold, the pixel is changed to a second predefined substitute value.

10. The method as recited in claim 9 , wherein:

a statistic is ascertained as to how each pixel of the input image or of the intermediate product varies over a predefined calibration set of input images;

a maximum pixel value of the statistic is established as the first predefined substitute value for the pixel; and

a minimum pixel value of the statistic is established as the second predefined substitute value for the pixel.

11. The method as recited in claim 1 , wherein a sum of the binary masks, which are each weighted with the assessments of the binary masks by the quality function, is ascertained as a decision-relevant component of the input image or of the intermediate product.

12. A method for measuring components of an input image on which an image classifier bases its decision regarding an assignment of the input image to one or multiple class(es) of a predefined classification, the method comprising the following steps:

providing binary masks which indicate which pixels of the input image and/or of an intermediate product formed in the image classifier, are considered relevant;

assessing each binary mask of the binary masks using a quality function, which is a measure of an extent to which at least one classification score, supplied by the image classifier, with respect to at least one target class changes when the pixels of the input image or of the intermediate product, which are relevant according to the binary mask, are changed; and

ascertaining sought-after components of the input image relevant for the decision of the image classifier from a combination of the binary masks with their respective assessments by the quality function, wherein a decision-relevant component of the intermediate product is converted into the sought-after component of the input image by interpolation or other upsampling.

13. A method for measuring components of an input image on which an image classifier bases its decision regarding an assignment of the input image to one or multiple class(es) of a predefined classification, the method comprising the following steps:

providing binary masks which indicate which pixels of the input image and/or of an intermediate product formed in the image classifier, are considered relevant;

assessing each binary mask of the binary masks using a quality function, which is a measure of an extent to which at least one classification score, supplied by the image classifier, with respect to at least one target class changes when the pixels of the input image or of the intermediate product, which are relevant according to the binary mask, are changed; and

ascertaining sought-after components of the input image relevant for the decision of the image classifier from a combination of the binary masks with their respective assessments by the quality function, wherein an image of a series-manufactured product is selected as the input image, and the classes of the classification represent a quality assessment of the product.

14. The method as recited in claim 13 , wherein the ascertained sought-after components of the input image on which the image classifier bases its decision are compared to a component of the input image which was ascertained to be relevant for the quality assessment of the product based on an observation of the same product using a different mapping modality, a quality assessment for the image classifier being ascertained from a result of the comparison.

15. A method for measuring components of an input image on which an image classifier bases its decision regarding an assignment of the input image to one or multiple class(es) of a predefined classification, the method comprising the following steps:

providing binary masks which indicate which pixels of the input image and/or of an intermediate product formed in the image classifier, are considered relevant;

assessing each binary mask of the binary masks using a quality function, which is a measure of an extent to which at least one classification score, supplied by the image classifier, with respect to at least one target class changes when the pixels of the input image or of the intermediate product, which are relevant according to the binary mask, are changed; and

ascertaining sought-after components of the input image relevant for the decision of the image classifier from a combination of the binary masks with their respective assessments by the quality function, wherein an image of a traffic situation recorded from a vehicle is selected as the input image, the classes of the classification representing assessments of the traffic situation, based on which a future behavior of the vehicle is planned.

16. The method as recited in claim 15 , wherein the ascertained sought-after components of the input image on which the image classifier bases its decision are compared to a component of the input image which is relevant for the assessment of the traffic situation, a quality assessment for the image classifier being ascertained from a result of this comparison.

17. A non-transitory machine-readable data medium on which is stored a computer program including machine-readable instructions for measuring components of an input image on which an image classifier bases its decision regarding an assignment of the input image to one or multiple class(es) of a predefined classification, the machine-readable instructions, when executed by one or more computers, causing the one or more computers to perform the following steps:

providing binary masks which indicate which pixels of the input image and/or of an intermediate product formed in the image classifier, are considered relevant;

assessing each binary mask of the binary masks using a quality function, which is a measure of an extent to which at least one classification score, supplied by the image classifier, with respect to at least one target class changes when the pixels of the input image or of the intermediate product, which are relevant according to the binary mask, are changed; and

ascertaining sought-after components of the input image relevant for the decision of the image classifier from a combination of the binary masks with their respective assessments by the quality function, wherein at least one derivative and/or one gradient of the at least one classification score with respect to the pixels of the input image or of the intermediate product which are relevant according to the binary mask has an improving effect on a value of the quality function.

18. A computer configured to measure components of an input image on which an image classifier bases its decision regarding an assignment of the input image to one or multiple class(es) of a predefined classification, the computer configured to:

provide binary masks which indicate which pixels of the input image and/or of an intermediate product formed in the image classifier, are considered relevant;

assess each binary mask of the binary masks using a quality function, which is a measure of an extent to which at least one classification score, supplied by the image classifier, with respect to at least one target class changes when the pixels of the input image or of the intermediate product, which are relevant according to the binary mask, are changed; and

ascertain sought-after components of the input image relevant for the decision of the image classifier from a combination of the binary masks with their respective assessments by the quality function, wherein at least one derivative and/or one gradient of the at least one classification score with respect to the pixels of the input image or of the intermediate product which are relevant according to the binary mask has an improving effect on a value of the quality function.