Title:
MODEL SELECTION METHOD AND IMAGE PROCESSING METHOD
Document Type and Number:
WIPO Patent Application WO/2023/162394
Kind Code:
A1
Abstract:
This model selection method for selecting a plurality of machine learning models for performing image processing comprises: a first model selection step (S33); and a second model selection step (S35). In the first model selection step (S33), by using a first image group that includes at least one confirmation image, a plurality of machine learning models are selected from all the machine learning models to form a first model group. In the second model selection step (S35), by using a second image group, at least one machine learning model is selected from the machine learning models included in the first model group. The number of confirmation images included in the second image group is greater than that of the first image group. By increasing the number of confirmation images in a stepwise manner and narrowing down the number of models to be selected, a highly accurate and appropriate machine learning model can be selected without using Ground Truth, from among a large number of machine learning models.
More Like This:
Inventors:
HIRAI MAKI (JP)
TOKISUE SHOGO (JP)
TOKISUE SHOGO (JP)
Application Number:
PCT/JP2022/044930
Publication Date:
August 31, 2023
Filing Date:
December 06, 2022
Export Citation:
Assignee:
SCREEN HOLDINGS CO LTD (JP)
International Classes:
G06T7/00; G01N33/48; G06T7/10
Domestic Patent References:
| WO2021095256A1 | 2021-05-20 | |||
| WO2015177268A1 | 2015-11-26 |
Foreign References:
| JP2021506022A | 2021-02-18 |
Attorney, Agent or Firm:
NISHIDA Takami (JP)
Download PDF: