The Ray Tracing based Tool for Generation Artificial Images and Neural Network Training

Alexey Kolker, Sofia Oshchepkova, Zhanna Pershina, Lubomir Dimitrov, Vladislav Ivanov, Aquib Rashid, Mohamad Bdiwi

Abstract

Creating quality-annotated dataset is one of the main tasks in the field of deep learning technologies for pattern recognition. However, in the real world, collecting a sufficient number of detailed images of an object is difficult and time-consuming. The article considers an approach to creating synthetic datasets based on the ray tracing method. This paper also presents the results of success tests of real object image segmentation by convolutional neural networks, trained entirely on synthetic data and data of different nature.

Download


Paper Citation