The experience of using artificial neural networks in solving problems of virtual reconstruction of historical manor interiors

The article discusses the experience of using artificial neural networks for virtual 3D reconstruction of historical interiors, using the example of the Kuzminki estate from the XVIII to early XX centuries. The Kuzminki manor complex, located in the southeastern part of Moscow, is a unique architectural monument and cultural heritage, with a history dating back more than 300 years. This estate was originally a summer residence of the Stroganov family and later passed into the hands of the Golitsyn family. Famous architects such as M.F. Kazakov, I.V. Egotov, A.N. Voronikhin, and representatives of the Gilardi family contributed to its construction. Bykovsky also participated in its development since the 1840s. In this paper, we demonstrate a hybrid approach that combines classical 3D modeling with neural network tools using the example of the reconstruction of the Kuzminki Manor ballroom. The approach involves the use of 3ds Max and Corona Renderer for 3D modeling, as well as Tripo AI, Prome AI, and Midjourney for neural network tools. Special attention is given to the reconstruction of an Empire-style ceremonial chair. This includes the generation of a 3D model based on 2D references, the creation of authentic textures, and the integration of the object into a virtual space. The relevance of this research lies in the need to develop effective methods for recreating lost cultural heritage sites using modern artificial intelligence (AI) technologies. The author of this study has been the first to explore the possibilities of utilizing artificial neural networks (ANNs) in solving the problem of virtual reconstruction of historical manor interiors, using the example of the Kuzminki estate. A comparison between classical and ANN methods has shown that ANNs offer new opportunities for creating virtual 3D interior reconstructions. They allow for a different approach to visualizing specific interior furnishings, which is not inferior to traditional modeling techniques. At the same time, ANNs act as a tool or virtual assistant, and the results can be controlled. The study confirmed the effectiveness of neural network technologies as a tool for the reconstruction of historic interiors. This is especially evident when using classical modeling and visualization techniques in conjunction with neural networks.