Learning device, learning method, and storage medium

A learning method of training a machine learning model on the basis of training data including correspondence between pixels of one or more images and correct-answer classes, the machine learning model being a convolutional neural network, receiving an input of an image including a plurality of pixels and outputting a degree of accuracy with which each pixel corresponds to a class indicating a type of an object, the learning method being performed by a computer, the learning method comprising:

• adjusting an output value of the degree of accuracy using a predetermined parameter with a tendency to decrease the output value of the degree of accuracy corresponding to a correct-answer class and to increase the output value of the degree of accuracy corresponding to a class other than the correct-answer class;
• and
• training the machine learning model on the basis of the adjusted output value, wherein the computer sets the same predetermined parameter for two or more classes between which a difference in the number of pieces of training data used as the correct-answer class for learning is determined to be equal to or less than a threshold value out of a plurality of classes.

A non-transitory computer-readable storage medium storing a program, the program causing a computer to train a machine learning model on the basis of training data including correspondence between pixels of one or more images and correct-answer classes, the machine learning model being a convolutional neural network, receiving an input of an image including a plurality of pixels and outputting a degree of accuracy with which each pixel corresponds to a class indicating a type of an object, the program causing the computer to perform:

• adjusting an output value of the degree of accuracy using a predetermined parameter with a tendency to decrease the output value of the degree of accuracy corresponding to a correct-answer class and to increase the output value of the degree of accuracy corresponding to a class other than the correct-answer class;
• and
• training the machine learning model on the basis of the adjusted output value, wherein the computer sets the same predetermined parameter for two or more classes between which a difference in the number of pieces of training data used as the correct-answer class for learning is determined to be equal to or less than a threshold value out of a plurality of classes.