Learning apparatus, moving image generating apparatus, method of generating learned model, moving image generating method, and program
AssigneeSony Interactive Entertainment Inc.
InventorHirotaka Asayama
Life of this patent
Estimating the term…
Abstract
A learning apparatus, a method of generating a learned model, and a program are capable of adequately widening the dynamic ranges of various images in a unified fashion. A training data generating section generates, on the basis of a second-class image, a first-class image associated with the second-class image, by referring to associative data where luminance values in a second dynamic range and luminance values in a first dynamic range are associated with each other. A learning section performs a learning process of a machine learning model by using the first-class image and the second-class image associated with the first-class image.
Claims20
A moving image generating apparatus comprising:
- a processor;
- and
- a storage unit storing programs executable by the processor to perform operations comprising:
- generating, based on a first-class video of a first dynamic range, a second-class video, the second-class video being a version of the first-class video with a second dynamic range into which the first dynamic range has been widened, wherein the generating the second-class video comprises:
- executing a forward generation process that, by using a learned model and on a basis of a first-class frame image that is a frame of the first-class video and a second-class frame image of a frame immediately before the first-class frame image, generates a second-class frame image of the first-class frame image, the second-class frame image of the frame immediately before the first-class frame image being generated by the learned model;
- executing a backward generation process that, by using the learned model and on a basis of the first-class frame image and a second-class frame image of a frame immediately after the first-class frame image, generates a second-class frame image of the first-class frame image, the second-class frame image of the frame immediately after the first-class video being generated by the learned model;
- and
- generating the second-class video using the second-class frame image generated by the forward generation process and the second-class frame image generated by the backward generation process.
A moving image generating method comprising:
- generating, based on a first-class video of a first dynamic range, a second-class video, the second-class video being a version of the first-class video with a second dynamic range into which the first dynamic range has been widened, wherein the generating the second-class video comprises:
- executing a forward generation process that, by using a learned model and on a basis of a first-class frame image that is a frame of the first-class video and a second-class frame image of a frame immediately before the first-class frame image, generates a second-class frame image of the first-class frame image, the second-class frame image of the frame immediately before the first-class frame image being generated by the learned model;
- executing a backward generation process that, by using the learned model and on a basis of the first-class frame image and a second-class frame image of a frame immediately after the first-class frame image, generates a second-class frame image of the first-class frame image, the second-class frame image of the frame immediately after the first-class frame image being generated by the learned model;
- and
- generating the second-class video using the second-class frame image generated by the forward generation process and the second-class frame image generated by the backward generation process.
A non-transitory, computer-readable storage medium containing a computer program, which when executed by a computer causes the computer to carry out actions comprising:
- generating, based on a first-class video of a first dynamic range, a second-class video, the second-class video being a version of the first-class video with a second dynamic range into which the first dynamic range has been widened, wherein the generating the second-class video comprises:
- executing a forward generation process that, by using a learned model and on a basis of a first-class frame image that is a frame of the first-class video and a second-class frame image of a frame immediately before the first-class frame image, generates a second-class frame image of the first-class frame image, the second-class frame image of the frame immediately before the first-class frame image being generated by the learned model;
- executing a backward generation process that, by using the learned model and on a basis of the first-class frame image and a second-class frame image of a frame immediately after the first-class frame image, generates a second-class frame image of the first-class frame image, the second-class frame image of the frame immediately after the first-class frame image being generated by the learned model;
- and
- generating the second-class video using the second-class frame image generated by the forward generation process and the second-class frame image generated by the backward generation process.
Dependencies are inferred from the claim text.