Homepage Program Systems: Theory and Applications Русская версия
ISSN 2079-3316 Bilingual online scientific Online scientific journal of the Ailamazyan Program System Institute of the Ailamazyan PSI of PSI of Russian Academy of Science of RAS 12+ 
Volume 17 (2026) . Issue 2 (71) . Paper No. 5 (512)

Artificial intelligence and machine learning

Research Article
DOIDigital Object Identifier 10.25209/2079-3316-2026-17-2-191-262

Roof-DeGAN: a hybrid GAN with cross-scale attention for aerial roof inpainting

Igor Victorovich Vinokurov1Correspondent author, Georgy Mikhailovich Lapankov2, Georgy Dmitrievich Umarov3

1-3Financial University under the Government of the Russian Federation, Moscow, Russia
1 Igor Victorovich Vinokurov — Correspondent author igvvinokurov@fa.ru

Abstract. This paper proposes a hybrid generative adversarial network, Roof-DeGAN, for restoring damaged and missing roof areas in aerial images. The architecture combines a densely connected Vision Transformer in the generator with a multi-scale discriminator featuring cross-scale attention. The model integrates the advantages of GANs, diffusion modeling elements, and transformer mechanisms. Experiments on real data from the PLC «Roscadastr» demonstrate that Roof-DeGAN outperforms existing methods, achieving PSNR = 33.7 dB, SSIM = 0.971, LPIPS = 0.048, and FID = 17.8 with an inference time of 0.15 seconds per 256×256 image. The developed approach shows high practical value for cadastre maintenance and cartographic data updating tasks. (Linked article texts in English and in Russian).

Keywords: generative adversarial networks, Roof-DeGAN, image inpainting, aerial imagery, remote sensing, roof reconstruction

MSC-20202020 Mathematics Subject Classification 68T20; 68T07, 68T45MSC-2020 68-XX: Computer science MSC-2020 68Txx: Artificial intelligence MSC-2020 68T20: Problem solving in the context of artificial intelligence (heuristics, search strategies, etc.) MSC-2020 68T07: Artificial neural networks and deep learning MSC-2020 68T45: Machine vision and scene understanding

For citation: Igor V. Vinokurov, Georgy M. Lapankov, Georgy D. Umarov. Roof-DeGAN: a hybrid GAN with cross-scale attention for aerial roof inpainting. Program Systems: Theory and Applications, 2026, 17:2, pp. 191–262. (in Engl. In Russ.). https://psta.psiras.ru/2026/2_191-262.

Full text of bilingual article (PDF): https://psta.psiras.ru/read/psta2026_2_191-262.pdf (Clicking on the flag in the header switches the page language).

The article was submitted 30.04.2026; approved after reviewing 01.06.2026; accepted for publication 10.06.2026; published online 20.06.2026.

© Vinokurov I. V., Lapankov G. M., Umarov G. D.
2026
Editorial address: Ailamazyan Program Systems Institute of the Russian Academy of Sciences, Peter the First Street 4«a», Veskovo village, Pereslavl area, Yaroslavl region, 152021 Russia;   Website:  http://psta.psiras.ru Phone: +7(4852) 695-228;   E-mail: ;   License: CC-BY-4.0License text on the Creative Commons site
© Ailamazyan Program System Institute of Russian Academy of Science (site design) 2010–2026