Robert Boschgmbh

Kanil Patel

Dan Zhang

Kilian Rambach

Michael Pfeiffer

William Harris Beluch

Liangyu Zhong

A method of training an image classifier. The image classifier determines a prediction vector of class probabilities for an input image. First, one or more initial training steps are performed, resulting in current values for the trainable parameters of the image classifier. A training image and corresponding class label are selected, and a current label vector is determined for training the image classifier on the training image. To this end, the image classifier is applied to the training image according to the current parameter values. If the class predicted by the image classifier is equal to the class label, the current label vector is determined as a soft label based on the prediction vector. A current training step of training the image classifier on the training image is performed using the current label vector.

On-the-fly calibration of an image classifier

A computer-implemented method of calibrating a trained classification model. The trained classification model is trained to classify input samples according to a plurality of classes and to provide associated prediction probabilities, and includes a plurality of hidden layers and at least one activation layer. The method includes accessing the trained classification model and accessing a plurality of validation samples, each validation sample having a ground-truth label, the ground-truth label indicating a ground-truth class. The method further includes applying the trained classification model to the plurality of validation samples, obtaining, for each validation sample, an output logit vector from a layer of the trained classification model preceding a last activation layer, and training a calibration module. The calibration module is trained to adjust prediction probabilities, the prediction probabilities being derived from the output logit vectors.

Classification model calibration

A method for training a classifier and/or regressor. The method includes: providing training samples labelled with ground truth classification and/or regression scores; and for each training sample from at least a subset of the training samples: determining a confidence score that quantifies an uncertainty of the training sample, and/or an ease or difficulty of classifying this sample, and reducing a largest ground truth classification and/or regression score with respect to one class and/or regression value by an amount that is dependent on the confidence score, and distributing the amount to ground truth classification and/or regression scores with respect to other classes and/or regression values; mapping the training samples to classification and/or regression scores; rating a deviation of the classification and/or regression scores from the updated ground truth classification and/or regression scores; and optimizing parameters that characterize the behavior of the classifier and/or regressor.

Training of classifiers and/or regressors on uncertain training data

A method is described for locating and/or classifying at least one object, a radar sensor that is used including at least one transmitter and at least one receiver for radar waves. The method includes: the signal recorded by the receiver is converted into a two- or multidimensional frequency representation; at least a portion of the two- or multidimensional frequency representation is supplied as an input to an artificial neural network, ANN that includes a sequence of layers with neurons, at least one layer of the ANN being additionally supplied with a piece of dimensioning information which characterizes the size and/or absolute position of objects detected in the portion of the two- or multidimensional frequency representation; the locating and/or the classification of the object is taken from the ANN as an output.

Growing community of inventors

Stuttgart, Germany

Kanil Patel