Information processing device, information processing method, and recording medium
View Patent ↗An information processing device includes a processor. The processor obtains a first classification threshold for classifying data into at least one of a plurality of classes, and outputs a classification result of classifying the data into at least one of the plurality of classes based on an output of a trained classification model and the first classification threshold. The first classification threshold is obtained by a second transform performed on a second classification threshold, the second transform being an inverse transform of a first transform. The first transform corresponds to transforming the output of the trained classification model into a classification probability value of each of a plurality of unit classes constituting the plurality of classes. The second classification threshold is set based on the classification probability values of the plurality of unit classes.
1. An information processing system comprising:
an information processing device including a first processor; and
a threshold calculating device including a second processor,
wherein the first processor is configured to:
obtain an image captured by a camera;
obtain a plurality of first classification thresholds outputted from the threshold calculating device, the plurality of first classification thresholds being a plurality of classification thresholds that are for classifying an object included in the image into at least one of a plurality of classes and correspond to the plurality of classes;
compare a plurality of first scalars and the plurality of first classification thresholds, wherein the plurality of first scalars are obtained by inputting the image into a first neural network that is a trained first classification model, wherein the plurality of first scalars correspond to the plurality of classes, wherein the plurality of first scalars are a plurality of scalars that are not transformed into classification probability values, and wherein the plurality of first classification thresholds correspond to the plurality of classes;
classify the object into, among the plurality of classes, at least one class having a corresponding first scalar that is greater than a corresponding first classification threshold; and
output, as a classification result, an information relating to the at least one class into which the object is classified,
wherein the second processor is configured to:
obtain a set of images for training that includes a plurality of images for training and a plurality of correct data items corresponding to the plurality of images for training;
train a second neural network using the image set for training, to output, from each of the plurality of images for training inputted, the plurality of first scalars corresponding to the plurality of classes, the second neural network being a classification model;
perform a first transform that transforms a plurality of second scalars into a plurality of classification probability values, the plurality of second scalars corresponding to the plurality of classes and being obtained by inputting an image for evaluation into the second neural network trained;
determine a plurality of second classification thresholds based on the plurality of classification probability values;
perform a second transform that is a transform from the plurality of second classification thresholds into the plurality of first classification thresholds and an inverse transform of the first transform; and
output the plurality of first classification thresholds into which the plurality of second classification thresholds have been transformed by the second transform to the information processing device.
2. The information processing system according to claim 1 ,
wherein the first transform is a computation by an inverse-transformable probability function, and
the second transform is a computation by an inverse function of the inverse-transformable probability function.
3. The information processing system according to claim 1 ,
wherein the first transform is a transform using a database corresponding to a computation by an inverse-transformable probability function, and
the second transform is a transform using a database corresponding to a computation by an inverse function of the inverse-transformable probability function.
4. The information processing system according to claim 1 ,
wherein the image for evaluation is a set of images for evaluation including a plurality of images for evaluation,
the plurality of images for evaluation are inputted into the second neural network trained, to obtain, for each of the plurality of images for evaluation, a plurality of classification probability values corresponding to the plurality of classes, and
the plurality of second classification thresholds are determined based on a classification result using the plurality of second classification thresholds based on the plurality of classification probability values obtained for each of the plurality of images for evaluation.
5. The information processing system according to claim 1 ,
wherein
the plurality of classes include a class corresponding to a person, a class corresponding to a vehicle, and a class different from the class corresponding to the person and the class corresponding to the vehicle.
6. An information processing method,
wherein the information processing method causes a first processor of an information processing device to:
obtain an image captured by a camera;
obtain a plurality of first classification thresholds outputted from a threshold calculating device including a second processor, the plurality of first classification thresholds being a plurality of classification thresholds that are for classifying an object included in the image into at least one of a plurality of classes and correspond to the plurality of classes;
compare a plurality of first scalars and the plurality of first classification thresholds, wherein the plurality of first scalars are obtained by inputting the image into a first neural network that is a trained first classification model, wherein the plurality of first scalars correspond to the plurality of classes, wherein the plurality of first scalars are a plurality of scalars that are not transformed into classification probability values, and wherein the plurality of first classification thresholds correspond to the plurality of classes;
classify the object into, among the plurality of classes, at least one class having a corresponding first scalar that is greater than a corresponding first classification threshold; and
output, as a classification result, an information relating to the at least one class into which the object is classified,
wherein the information processing method further causes the second processor to:
obtain a set of images for training that includes a plurality of images for training and a plurality of correct data items corresponding to the plurality of images for training;
train a second neural network using the image set for training, to output, from each of the plurality of images for training inputted, the plurality of first scalars corresponding to the plurality of classes, the second neural network being a classification model;
perform a first transform that transforms a plurality of second scalars into a plurality of classification probability values, the plurality of second scalars corresponding to the plurality of classes and being obtained by inputting an image for evaluation into the second neural network trained;
determine a plurality of second classification thresholds based on the plurality of classification probability values;
perform a second transform that is a transform from the plurality of second classification thresholds into the plurality of first classification thresholds and an inverse transform of the first transform; and
output the plurality of first classification thresholds into which the plurality of second classification thresholds have been transformed by the second transform to the information processing device.