Features of Preparing a Data Set for Training a Neural Network of an Object Tracking System in a Stream

This article is devoted to discussing the influence of the content of a data set and the features of its markup on the quality of the resulting neural network. To perform the tasks of detecting and localizing objects, the YOLOv8 neural network is used. A brief motivation for using this particular neural network with appropriate links is given. This network is trained on three similar data sets, after which the results obtained are compared and analyzed regarding the real problem. In conclusion, general recommendations are given for compiling a data set for training a neural network, as well as suggestions for the further development of this work.