Learning data generating apparatus, learning data generating method, and non-transitory recording medium having learning data generating program recorded thereon

A learning data generating method performed with a learning data generating apparatus, the learning data generating apparatus including an image input processor comprising hardware, a first neural network application processor comprising hardware, a second neural network application processor comprising hardware, a correct answer position input processor comprising hardware, and an output checking processor comprising hardware, the method comprising:

• acquiring, with the image input processor, first training data obtained by picking up a still image from a series of images and first correct answer information that is position information of a recognition target in the first training data;
• acquiring, with the image input processor, second training data obtained by picking up another still image from the series of images;
• inputting, with the first neural network application processor, the first training data into a neural network to thereby cause the neural network to output a first array group constituted of a plurality of vectors;
• inputting, with the second neural network application processor, the second training data into a neural network to thereby cause the neural network to output a second array group constituted of a plurality of vectors;
• setting, as a first array, with the correct answer position input processor, a vector at a position indicated by the first correct answer information in the first array group;
• and
• determining, with the output checking processor, a difference between the first array and each of the plurality of vectors constituting the second array group, to create second correct answer information in the second training data based on the difference, the second correct answer information being position information of the second training data.

A non-transitory recording medium having a learning data generating program recorded thereon, the program being configured to cause a learning data generating apparatus to:

• acquire first training data obtained by picking up a still image from a series of images and first correct answer information that is position information of a recognition target in the first training data;
• acquire second training data obtained by picking up another still image from the series of images;
• input the first training data into a neural network to thereby cause the neural network to output a first array group constituted of a plurality of vectors;
• input the second training data into a neural network to thereby cause the neural network to output a second array group constituted of a plurality of vectors;
• set, as a first array, a vector at a position indicated by the first correct answer information in the first array group;
• and
• determine a difference between the first array and each of the plurality of arrays vectors constituting the second array group to create second correct answer information in the second training data based on the difference, the second correct answer information being position information of the second training data.