Image generation device, image generation method, and image generation program
An image generation means 81 generates an image using a generator. A discrimination means 82 discriminates whether an object image includes a feature of a target image, using a discriminator. A first update means 83 updates the generator so as to minimize a first error representing a degree of divergence between a result of discriminating a generated image using the discriminator and a correct answer label associated with the generated image, the generated image being the image generated using the generator. A second update means 84 updates the discriminator so as to minimize a second error representing a degree of divergence between each of respective results of discriminating the generated image, a first actual image including the feature of the target image, and a second actual image not including the feature of the target image using the discriminator and a correct answer label associated with a corresponding image.
1. An image generation device comprising a hardware processor configured to execute software code to:
generate an image using a generator;
discriminate whether an object image includes a feature of a target image, using a discriminator;
discriminate the generated image, using the discriminator, yielding a first result;
discriminate a first actual image that includes the feature of the target image, using the discriminator, yielding a second result;
discriminate a second actual image that does not include the feature of the target image, using the discriminator, yielding a third result;
update the generator so as to minimize a first error representing a first degree of divergence between the first result and a first correct answer label associated with the generated image and indicating that the generated image is not the target image;
update the discriminator so as to minimize a second error representing a second degree of divergence between the first result and the first correct answer label, between the second result and a second correct answer label associated with the first actual image and indicating that the first actual image is the target image, and between the third result and a third correct answer label associated with the second actual image and indicating that the second actual image is not the target image;
output the first result and the first correct answer label, in a first case in which the object image is the generated image;
output the second result and the second correct answer label in a second case in which the object image is the first actual image; and
output the third result and the third correct answer label, in a third case in which the object image is the second actual image.
2. The image generation device according to claim 1 , wherein the hardware processor is configured to execute the software code to update the generator so as to minimize the first error representing the first degree of divergence in the first case in which the object image is the generated image.
3. The image generation device according to claim 1 , wherein the hardware processor is configured to execute the software code to:
update the generator so as to minimize the first error between the first result and a correct answer that is the first correct answer in the first case, the second correct answer in the second case, and the third correct answer in the third case; and
update the discriminator so as to minimize the second error between each of the first result, the second result, and the third result and the correct answer that is the first correct answer in the first case, the second correct answer in the second case, and the third correct answer in the third case.
4. The image generation device according to claim 1 , wherein the hardware processor is configured to execute the software code:
output 0 as the first correct answer label in the first case;
output 1 as the second correct answer label in the second case; and
output 0 as the third correct answer label in the third case.
5. An image generation method comprising:
generating an image using a generator;
discriminating whether an object image includes a feature of a target image, using a discriminator;
discriminating the generated image, using the discriminator, yielding a first result;
discriminating a first actual image that includes the feature of the target image, using the discriminator, yielding a second result;
discriminating a second actual image that does not include the feature of the target image, using the discriminator, yielding a third result;
updating the generator so as to minimize a first error representing a first degree of divergence between the first result and a first correct answer label associated with the generated image and indicating that the generated image is not the target image;
updating the discriminator so as to minimize a second error representing a second degree of divergence between the first result and the first correct answer label, between the second result and a second correct answer label associated with the first actual image and indicating that the first actual image is the target image, and between the third result and a third correct answer label associated with the second actual image and indicating that the second actual image is not the target image;
outputting the first result and the first correct answer label, in a first case in which the object image is the generated image;
outputting the second result and the second correct answer label in a second case in which the object image is the first actual image; and
outputting the third result and the third correct answer label, in a third case in which the object image is the second actual image.
6. A non-transitory computer readable information recording medium storing an image generation program, when executed by a processor, that performs a method for:
generating an image using a generator;
discriminating whether an object image includes a feature of a target image, using a discriminator;
discriminating the generated image, using the discriminator, yielding a first result;
discriminating a first actual image that includes the feature of the target image, using the discriminator, yielding a second result;
discriminating a second actual image that does not include the feature of the target image, using the discriminator, yielding a third result;
updating the generator so as to minimize a first error representing a first degree of divergence between the first result and a first correct answer label associated with the generated image and indicating that the generated image is not the target image;
updating the discriminator so as to minimize a second error representing a second degree of divergence between the first result and the first correct answer label, between the second result and a second correct answer label associated with the first actual image and indicating that the first actual image is the target image, and between the third result and a third correct answer label associated with the second actual image and indicating that the second actual image is not the target image;
outputting the first result and the first correct answer label, in a first case in which an object image is the generated image;
outputting the second result and the second correct answer label in a second case in which the object image is the first actual image; and
outputting the third result and the third correct answer label, in a third case in which the object image is the second actual image.