DATACET ANNOTATION METHOD IN ABSENCE OF INITIAL MARKING
K. A. Maykov, L.Е. Tartykov, Sergey Alexeevich Bubnov · 2025
A method for automatic dataset marking is proposed in the absence of initial marking and a pre-trained neural network model. At the initial stage, pseudo-marks are generated, which are corrected by the expert at a low level of the confidence threshold. To reduce the error, the neural network is retrained on reliable results selected independently, taking into account the manual corrections made by the expert. The method allows you to work in a narrowly focused subject area