This article explores the development and implementation of mathematical models and deep learning methods for reconstructing a 3D scene from 2D images. In this area of research, the use of neural networks for reconstructing 3D scenes based on 2D images is actively developing. This approach is used in various fields, including robotics, virtual reality, and the creation of 3D models of objects. Neural networks are trained on large data sets containing pairs of 2D images and corresponding 3D scenes. During training, the network captures the complex relationships between image pixels and the depth of objects in the scene. The purpose of using this system is to create a 3D model of a real scene using only 2D images taken from different angles.