Granted Patent B2

US 12,243,283 · App. 17/894,358 · Granted Mar 4, 2025

Inventors: Chaithanya Kumar Mummadi (Pittsburgh, PA); Jan Hendrik Metzen (Boeblingen, DE); Robin Hutmacher (Renningen, DE)

Assignee: ROBERT BOSCH GMBH

G06V10/26G06V10/764G06V10/7715G06V10/82

Abstract

Claims (40)

1. A computer-implemented method for determining an output signal characterizing a semantic segmentation and/or an instance segmentation of an image, the method comprising the following steps:

determining a first intermediate output signal from a machine learning system, the first intermediate output signal characterizing a semantic segmentation and/or an instance segmentation of the image;

adapting parameters of the machine learning system based on a loss function, wherein the loss function characterizes an entropy or a cross-entropy of the first intermediate output signal; and

determining the output signal from the machine learning system based on the image and the adapted parameters;

wherein the loss function characterizes a mean entropy of classifications obtained for pixels of the image;

wherein a second intermediate output signal is determined based on the first intermediate output signal by using a transformation function,

wherein the second intermediate output signal characterizes a semantic segmentation and/or an instance segmentation of the image and the cross-entropy is determined based on the first intermediate output signal and the second intermediate output signal; and

wherein the transformation function characterizes an edge-preserving smoothing filter.

2. The method according to claim 1 , wherein the loss function further characterizes a likelihood of at least a part of the first intermediate output signal, wherein the likelihood is determined based on a density model of the at least part of the image.

3. The method according to claim 1 , wherein the loss function further characterizes a log-likelihood, of at least a part of the first intermediate output signal, wherein the log-likelihood is determined based on a density model of the at least part of the image.

4. The method according to claim 2 , wherein the likelihood characterizes an average likelihood of a plurality of patches, wherein the plurality of patches is determined based on the first intermediate output signal.

5. The method according to claim 4 , wherein the likelihood of a patch of the patches is determined by determining a feature representation of the patch using a feature extractor and providing a likelihood of the feature representation as likelihood of the patch, wherein the likelihood of the feature representation is determined by the density model.

6. The method according to claim 2 , wherein the density model is characterized by a mixture model, or a Gaussian mixture model, or a normal distribution.

7. The method according to claim 1 , wherein for determining the output signal, the machine learning system includes a normalization transformation and the loss further characterizes a Kullback-Leibler divergence between an output of the normalization transformation and a predefined probability distribution.

8. The method according to claim 7 , wherein the predefined probability distribution is characterized by a standard normal distribution.

9. The method according to claim 7 , wherein the machine learning system includes a neural network for determining the output signal, wherein the neural network includes a normalization layer and wherein the loss further characterizes a Kullback-Leibler divergence between an output of the normalization layer and a predefined probability distribution.

10. A machine learning system configured to determine an output signal characterizing a semantic segmentation and/or an instance segmentation of an image, the machine learning system configured to:

determine a first intermediate output signal from the machine learning system, the first intermediate output signal characterizing a semantic segmentation and/or an instance segmentation of the image;

adapt parameters of the machine learning system based on a loss function, wherein the loss function characterizes an entropy or a cross-entropy of the first intermediate output signal; and

determine the output signal from the machine learning system based on the image and the adapted parameters;

wherein the loss function characterizes a mean entropy of classifications obtained for pixels of the image;

wherein a second intermediate output signal is determined based on the first intermediate output signal by using a transformation function,

wherein the second intermediate output signal characterizes a semantic segmentation and/or an instance segmentation of the image and the cross-entropy is determined based on the first intermediate output signal and the second intermediate output signal; and

wherein the transformation function characterizes an edge-preserving smoothing filter.

11. A control system configured to determine a control signal, wherein the control signal is configured to control an actuator and/or a display, and wherein the control signal is determined based on an output signal determined by:

determining a first intermediate output signal from a machine learning system, the first intermediate output signal characterizing a semantic segmentation and/or an instance segmentation of the image;

adapting parameters of the machine learning system based on a loss function, wherein the loss function characterizes an entropy or a cross-entropy of the first intermediate output signal; and

determining the output signal from the machine learning system based on the image and the adapted parameters;

wherein the loss function characterizes a mean entropy of classifications obtained for pixels of the image;

wherein a second intermediate output signal is determined based on the first intermediate output signal by using a transformation function,

wherein the second intermediate output signal characterizes a semantic segmentation and/or an instance segmentation of the image and the cross-entropy is determined based on the first intermediate output signal and the second intermediate output signal; and

wherein the transformation function characterizes an edge-preserving smoothing filter.

12. A non-transitory machine-readable storage medium on which is stored a computer program for determining an output signal characterizing a semantic segmentation and/or an instance segmentation of an image, the computer program, when executed by a computer, causing the computer to perform the following steps:

determining a first intermediate output signal from a machine learning system, the first intermediate output signal characterizing a semantic segmentation and/or an instance segmentation of the image;

adapting parameters of the machine learning system based on a loss function, wherein the loss function characterizes an entropy or a cross-entropy of the first intermediate output signal; and

determining the output signal from the machine learning system based on the image and the adapted parameters;

wherein the loss function characterizes a mean entropy of classifications obtained for pixels of the image;

wherein a second intermediate output signal is determined based on the first intermediate output signal by using a transformation function,

wherein the second intermediate output signal characterizes a semantic segmentation and/or an instance segmentation of the image and the cross-entropy is determined based on the first intermediate output signal and the second intermediate output signal; and

wherein the transformation function characterizes an edge-preserving smoothing filter.

Assignments (1)

Priority Claims (1)

Continuity (1)

References Cited (31)