Estimation system, estimation method, and recording medium recording estimation program

Patent metadata

Field	Value
Publication number	US-12412425-B2
Application number	US-202217852708-A
Country	US
Kind code	B2
Filing date	Jun 29, 2022
Priority date	Jul 21, 2021
Publication date	Sep 9, 2025
Grant date	Sep 9, 2025

Abstract

An estimation system includes: an acquisition processor that acquires a captured image of an object to be estimated; and an estimation processor that uses a single learned model generated on the basis of learning data, in which an image of the object to be estimated is associated with each of a plurality of attributes of the object to be estimated, to estimate a first attribute included in the plurality of attributes, from a first output value of a first output layer corresponding to the first attribute and to estimate a second attribute included in the plurality of attributes, from a second output value of a second output layer corresponding to the second attribute, with the captured image acquired by the acquisition processor as an input image.

First claim

The invention claimed is:

1. An estimation system comprising:
an acquisition circuit that acquires a captured image of an object to be estimated; and
an estimation circuit that uses a single learned model generated based on learning data, in which an image of the object to be estimated is associated with each of a plurality of attributes of the object to be estimated, to estimate a first attribute, included in the plurality of attributes, from a first output value of a first output layer corresponding to the first attribute and to estimate a second attribute, included in the plurality of attributes, from a second output value of a second output layer corresponding to the second attribute, with the captured image acquired by the acquisition circuit as an input image, wherein
the estimation circuit further estimates a candidate class based on the second output value calculated for each of a plurality of classes in which the second attribute is classified, and estimates the second attribute through linear interpolation using a plurality of candidate values included in the candidate class.

2. The estimation system according to claim 1 , wherein
the estimation circuit simultaneously estimates the first attribute and the second attribute.

3. The estimation system according to claim 1 , wherein
the estimation circuit calculates the first output value for each of a plurality of classifications of the first attribute, and estimates the first attribute for the input image based on a plurality of the calculated first output values.

4. The estimation system according to claim 1 , wherein
the estimation circuit outputs a total number of output values, which is a sum of a number of classifications of the first attribute and a number of classifications of the second attribute.

5. The estimation system according to claim 1 , wherein
the single learned model is generated based on learning data in which a face image of a person, a gender of the person, and an age of the person are associated with each other,
the acquisition circuit acquires the face image of the person, and
the estimation circuit uses the single learned model to estimate the gender of the person as the object to be estimated from the first output value of the first output layer corresponding to the gender of the person and to estimate the age of the person as the object to be estimated from the second output value of the second output layer corresponding to the age of the person, with the face image acquired by the acquisition circuit as the input image.

6. The estimation system according to claim 5 , wherein
the estimation circuit outputs a total number of output values, which is a sum of a number of gender classifications and a number of a plurality of classes of age classifications into which ages are classified.

7. The estimation system according to claim 6 , wherein
the estimation circuit estimates the age of the person as the object to be estimated based on i) a result of a product-sum operation of a plurality of second output values, including the second output value, associated with the plurality of classes of age classifications, and ii) age classes respectively corresponding to the plurality of second output values.

8. The estimation system according to claim 7 , wherein
the estimation circuit further estimates an age class, among the plurality of classes of age classifications, based on the result of the product-sum operation, and calculates an age corresponding to the second output value through linear interpolation using a minimum age and a maximum age, among a plurality of ages included in the estimated age class.

9. An estimation method performed by one or more processors, the estimation method comprising:
acquiring a captured image of an object to be estimated;
using a single learned model generated based on learning data, in which an image of the object to be estimated is associated with each of a plurality of attributes of the object to be estimated, to estimate a first attribute, included in the plurality of attributes, from a first output value of a first output layer corresponding to the first attribute and to estimate a second attribute, included in the plurality of attributes, from a second output value of a second output layer corresponding to the second attribute, with the acquired captured image as an input image; and
estimating a candidate class based on the second output value calculated for each of a plurality of classes in which the second attribute is classified, and estimating the second attribute through linear interpolation using a plurality of candidate values included in the candidate class.

10. A non-transitory computer-readable recording medium of a device coupled to at least one processor of the device and storing one or more computer-executable instructions for estimation, the one or more computer-executable instructions, when executed by the at least one processor, causing the device to:
acquire a captured image of an object to be estimated;
use a single learned model generated based on learning data, in which an image of the object to be estimated is associated with each of a plurality of attributes of the object to be estimated, to estimate a first attribute, included in the plurality of attributes, from a first output value of a first output layer corresponding to the first attribute and to estimate a second attribute, included in the plurality of attributes, from a second output value of a second output layer corresponding to the second attribute, with the acquired captured image as an input image; and
estimate a candidate class based on the second output value calculated for each of a plurality of classes in which the second attribute is classified, and estimate the second attribute through linear interpolation using a plurality of candidate values included in the candidate class.

Assignees

• Sharp Kk

Inventors

• Furuichi Gaku
• Ishida Hiroshi

Classifications

• G06V10/764

  using classification, e.g. of video objects · CPC title

• G06V10/82

  using neural networks · CPC title

• G06N3/04

  Architecture, e.g. interconnection topology · CPC title

• G06N3/09

  Supervised learning · CPC title

• G06N3/0464

  Convolutional networks [CNN, ConvNet] · CPC title