Device and method for training an image segmentation system

A computer-implemented method for determining an output signal, wherein the output signal characterizes a classification of a pixel of an input image as anomalous or not, wherein the method comprises the following steps:

• providing the input image as input to an image segmentation system, wherein the image segmentation system has been trained by:
• obtaining a training image and a training label map, wherein the training label map indicates pixels of the training image as belonging to an anomaly class,
• providing the training image as input to the image segmentation system, wherein the image segmentation system determines values characterizing logits for different classes for pixels of the training image,
• determining a loss value based on a loss function, wherein the loss function includes a first term that characterizes a logarithm of a sigmoid of logits determined for pixels in the training image, wherein the logarithm of the sigmoid is determined for pixels, which are labeled as belonging to the anomaly class in the training label map, and
• training the image segmentation system by minimizing the loss value;
• determining, by the image segmentation system, logits for the pixel;
• providing a classification of the pixel as anomalous in the output signal when a difference of a largest determined logit and a smallest determined logit is smaller or equal to a predefined threshold, and providing a classification of the pixel as non-anomalous in the output signal otherwise.

A training system, which is configured to train an image segmentation system, the training system configured to:

• obtain a training image and a training label map, wherein the training label map indicates pixels of the training image as belonging to an anomaly class;
• provide the training image as input to the image segmentation system, wherein the image segmentation system determines values characterizing logits for different classes for pixels of the training image;
• determine a loss value based on a loss function, wherein the loss function includes a first term that characterizes a logarithm of a sigmoid of logits determined for pixels in the training image, wherein the logarithm of the sigmoid is determined for pixels, which are labeled as belonging to the anomaly class in the training label map;
• and
• train the image segmentation system by minimizing the loss value.

A control system, which is configured to:

• determine an output signal, wherein the output signal characterizes a classification of a pixel of an input image as anomalous or not, the determining of the output signal including:
• providing the input image as input to an image segmentation system, wherein the image segmentation system has been trained by:
• obtaining a training image and a training label map, wherein the training label map indicates pixels of the training image as belonging to an anomaly class,
• providing the training image as input to the image segmentation system, wherein the image segmentation system determines values characterizing logits for different classes for pixels of the training image,
• determining a loss value based on a loss function, wherein the loss function includes a first term that characterizes a logarithm of a sigmoid of logits determined for pixels in the training image, wherein the logarithm of the sigmoid is determined for pixels, which are labeled as belonging to the anomaly class in the training label map, and
• training the image segmentation system by minimizing the loss value;
• determining, by the image segmentation system, logits for the pixel;
• providing a classification of the pixel as anomalous in the output signal when a difference of a largest determined logit and a smallest determined logit is smaller or equal to a predefined threshold, and providing a classification of the pixel as non-anomalous in the output signal otherwise;
• determine a control signal based on the output signal;
• control, using the control signal, an actuator and/or a display.

A non-transitory machine-readable storage medium on which is stored a computer program for training an image segmentation system, the computer program, when executed by a computer, causing the computer to perform the following steps:

• obtaining a training image and a training label map, wherein the training label map indicates pixels of the training image as belonging to an anomaly class;
• providing the training image as input to the image segmentation system, wherein the image segmentation system determines values characterizing logits for different classes for pixels of the training image;
• determining a loss value based on a loss function, wherein the loss function includes a first term that characterizes a logarithm of a sigmoid of logits determined for pixels in the training image, wherein the logarithm of the sigmoid is determined for pixels, which are labeled as belonging to the anomaly class in the training label map;
• and
• training the image segmentation system by minimizing the loss value.