AUTOMATION OF DATA MARKUP FOR NEURAL NETWORKS

Abstract

The process of digital transformation affects all spheres of life in modern society. All areas of human activity are now subject to digitalization in one form or another: economics, ecology, sci-entific activities, etc. Although some technologies, together with their accompanying competencies, become outdated, new technologies are emerging that require effective methods for recognizing elements of the environment, tracking the movements of controlled objects in interaction, including augmented reality. At the same time, neural networks are becoming increasingly popular for solving problems of recognizing objects in static images, as well as in video streams. The data markup for further deep learning is an integral part of any project that involves the use of a neural network of computer vision technologies. For training neural networks the task of classifying images or finding objects requires a huge amount of marked data. The more classes of objects, the more data is required for training. Naturally, time-consuming manual marking has a number of drawbacks: the long marking time itself and possible errors due to the routine of the task. The article discusses various approaches for automating the task of marking images for deep learning, which can be used to integrate image recognition technologies and create an augmented reality, with possible implementation on personal computers and mobile devices.