Learning method and recording medium

A non-transitory computer-readable recording medium storing a program which causes a computer to:

• input, to a neural network, a first image and a second image that constitute a moving image and that are temporally adjacent to each other, the second image being an image subsequent to the first image with a predetermined time interval therebetween;
• i) cause a first neural network that constitutes the neural network to use the first image and the second image and learn to output a first motion amount image representing a first amount of motion of each of pixels of the first image that continues until the predetermined time interval elapses and ii) cause a second neural network that constitutes the neural network and that differs from the first neural network to use the first image, the second image, and the first motion amount image and learn to output a second motion amount image representing a second amount of motion of each of the pixels of the first image that continues until the predetermined time interval elapses;
• and
• output the second motion amount image as a result of estimation of motion between the first image and the second image, wherein
• causing the first and second neural networks to learn to output the motion amount images involves causing the first neural network to use the first image and the second image and learn to output a transformation matrix applied to all the pixels of the first image and used to convert the first image into the second image and causing the first neural network to learn to output the first motion amount image generated from the transformation matrix,
• the transformation matrix carries out global transformation tracking the overall motion between the first and second images, but does not track the detailed motions of individual objects in the images,
• the first motion amount image generated using the transformation matrix represents the result of estimation of the overall motion between the first image and the second image, but does not represent estimation of the detailed motions of individual objects in the images, and
• the second motion amount image represents estimation of the detailed motions of individual objects in the images, wherein the program also causes the computer to
• a) generate a first estimated image from only the transformation matrix and the first image, the first estimated image being an image subsequent to the first image with the predetermined time interval therebetween,
• b) calculate a first error between the first estimated image and the second image,
• c) generate the second motion amount image from the transformation matrix and the first and second images,
• d) generate a second estimated image from the second motion amount image and the first image, the second estimated image being an image subsequent to the first image with the predetermined time interval therebetween,
• e) calculate a second error between the second estimated image and the second image,
• f) combine the first and second errors, and
• g) determine weights of the first neural network and the second neural network that minimize the combined error.