Loading [MathJax]/extensions/MathZoom.js
Development of semi-automatic image annotation using object recognition | IEEE Conference Publication | IEEE Xplore

Development of semi-automatic image annotation using object recognition


Abstract:

The modern field of machine learning has made remarkable progress, and the demand for training data required by machine learning systems is constantly increasing. However...Show More

Abstract:

The modern field of machine learning has made remarkable progress, and the demand for training data required by machine learning systems is constantly increasing. However, the annotation work required to create this training data requires a great deal of effort. Therefore, we considered semi-automating the labeling process by taking advantage of false positives that occur when object detection is performed by YOLOv5, thereby eliminating the need for region selection and reducing the cost of annotation work. In a comparison experiment with labelImg, a common annotation tool, we found that semi-automating the annotation process took less time than manually annotating the objects.
Date of Conference: 08-13 July 2023
Date Added to IEEE Xplore: 29 December 2023
ISBN Information:
Print on Demand(PoD) ISSN: 2472-0070
Conference Location: Koriyama, Japan

I. Introduction

The modern field of machine learning has made remarkable progress, and the demand for training data required by machine learning systems is constantly increasing. The quantity and quality of this training data must be of a high standard, and the annotation process requires a great deal of effort, as well as a high level of skill in its creation. In this paper, we first examine the feasibility of automating the annotation process, and show that it is difficult to fully automate the annotation process in practice. Next, we examine the reduction of annotation cost by semi-automation of annotation, design a semi-automatic annotation tool, and conduct experiments to confirm the effectiveness of the tool. Specifically, we will realize semi-automated annotation using the results of machine learning-based image recognition, and verify the effectiveness of our proposal by having university students perform annotation on images.

Contact IEEE to Subscribe

References

References is not available for this document.