Title:
IMAGE RECOGNITION SYSTEM
Document Type and Number:
WIPO Patent Application WO/2023/228722
Kind Code:
A1
Abstract:
An image recognition system according to one aspect of the present invention comprises: a discrepancy information extraction unit that accepts the input of inference results of conventional and new machine learning models for the same input image, and if there is a discrepancy area in the two inference results, outputs image information on the discrepancy area in the input image and information indicating whether the new machine learning model detects a point of interest in the discrepancy area; an object presence determination unit that determines whether a point of interest is included in the image information on the discrepancy area and outputs a determination result; and a degraded performance determination unit that determines that the performance of the new machine learning model is degraded in comparison to the current machine learning model on the basis of the information indicating whether the new machine learning model detects a point of interest in the discrepancy area and the determination result regarding whether a point of interest is present or absent in the image information on the discrepancy area.
Inventors:
KITAYAMA AKIRA (JP)
ONO GOICHI (JP)
ITO HIROAKI (JP)
ONO GOICHI (JP)
ITO HIROAKI (JP)
Application Number:
PCT/JP2023/017414
Publication Date:
November 30, 2023
Filing Date:
May 09, 2023
Export Citation:
Assignee:
HITACHI ASTEMO LTD (JP)
International Classes:
G06V10/776; G06N20/00; G06T7/00
Foreign References:
| JP2022061191A | 2022-04-18 | |||
| US20220237521A1 | 2022-07-28 |
Attorney, Agent or Firm:
SHIN-YU INTERNATIONAL PATENT FIRM (JP)
Download PDF: