Концептуальна модель доповнення точкових хмар на основі графових нейронних мереж

Abstract

A conceptual model for completing point clouds using graph neural networks has been developed, which allows to efficiently encode incomplete point clouds as graphs and predict missing points. The proposed approach allows for high-quality reconstruction of incomplete 3D data, which is extremely important for many applications such as 3D reconstruction, robot navigation, etc. Research methods are analysis and synthesis of scientific literature, systematic approach, machine learning methods, in particular graph neural networks. Further research may involve testing different neural network architectures, tuning hyperparameters, applying alternative loss functions, and using more powerful computing resources to train deep models.