Information processing method, information processing system, and information processing program

CN120917480APending Publication Date: 2025-11-07PANASONIC INTELLECTUAL PROPERTY MANAGEMENT CO LTD
-1 Cites -1 Cited by

Patent Information

Application Number
CN202480023019.X
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Priority Date
2023-04-13
Filing Date
2024-03-27
Publication Date
2025-11-07

AI Technical Summary

Technical Problem

Existing metrics for evaluating the inference accuracy of image recognition models, such as mIoU and mPA, fail to reflect the reliability of the inference results, making it difficult to accurately evaluate the accuracy of machine learning models.

Method used

The model accuracy is evaluated by acquiring the inference results and uncertainty information of the image recognition model, combining the positive solution labels to generate inference index information, and considering the confidence level of the inference results.

Benefits of technology

It can distinguish between accidental correct solutions and certain correct solutions, thus improving the accuracy of the inference accuracy evaluation of image recognition models.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN120917480A_ABST
    Figure CN120917480A_ABST
Patent Text Reader

Abstract

The evaluation device acquires a reasoning result of an image to be evaluated of an image recognition model generated by machine learning and uncertainty information indicating a degree of instability of the reasoning result, and calculates an evaluation result by using a positive solution tag associated with the image to be evaluated, the reasoning result, and the uncertainty information. And generating inference index information that imparts, to information indicating whether the inference result is a positive solution or a non-positive solution, a certainty degree that indicates which degree of the inference result is determined, and evaluating the inference accuracy of the image recognition model on the basis of the inference index information.
Need to check novelty before this filing date? Find Prior Art