11:57, 06 January 2026

Neural Networks in Russia Reconstruct Historic Monuments and “Read” Old Photos

New technologies help visualize lost details of cultural heritage sites and identify people in historical images.

Russian researchers have developed a neural-network-based method for restoring historical objects. Mikhail Babaytsev, a postgraduate researcher at the Department of the History of Philosophy and Theory of Culture at Tver State University and chief executive of the science-driven company Avrosystem, told IT RUSSIA about the work.

Together with his team, Babaytsev has spent several years digitizing monuments and artifacts. Using the data they collected, the researchers trained a neural network to generate elements of traditional wooden architecture.

“Training was carried out using real historical objects – photographs, drawings, and 3D models of preserved examples. The results were systematically reviewed by subject-matter experts. As a result, we developed an algorithm capable of generating typical wooden houses with a high degree of historical accuracy, based on official sources such as archival plans, descriptions, and construction practices of specific periods,” Babaytsev said.

The researchers are also applying machine learning to other tasks, including the automatic classification of architectural elements – distinguishing window frames, carved details, porches, and log structures.

A Catalyst for Scientific Research

“When it comes to analyzing damage or forecasting the deterioration of monuments, these are tasks for restorers, structural engineers, and cultural heritage preservation specialists. But artificial intelligence and neural networks could become powerful tools in this area as well – acting as a catalyst for analysis and prediction,” Babaytsev concluded.

Russian specialists have also trained neural networks to classify faces in historical photographs. Archives contain collections of black-and-white images from different years that lack captions. Visual inspection alone made it impossible to identify specific individuals.

“We trained a neural network using images of key figures we were interested in and then ran a search across the rest of the archive. The results were unexpected: the system found these people in photographs where we had not even assumed they were present. In some cases, it identified the same person in situations where we, as humans, were unsure,” the researcher said.

This clearly demonstrates how artificial intelligence can not only complement, but also extend the capabilities of historical research.

Culture, sports and media