The YandexART neural network helped restore video recordings of historical events, which found application in the new series "Games," dedicated to the preparation and organization of the 1980 Olympics. The neural network has learned to restore archival video recordings and newsreels, which is of significant value to film companies, museums, historians, and other specialists involved in preserving historical heritage.
YandexART was able to improve the quality of video footage from Mosfilm and the Russian State Archive of Film and Photo Documents. Thanks to this technology, modern viewers can see events from forty years ago in good quality. The neural network restored missing details, improved the texture of images, and increased the resolution and sharpness of video frames. For example, it added detail to the brickwork of walls, patterns on women's dresses, and details of cars that were poorly visible in the original recordings.
One of the features of YandexART is its ability to avoid "hallucinations" — situations where the neural network adds fictional elements. Thanks to step-by-step training on low-quality materials, the model can turn noisy and blurry frames into content with high detail.
Among the restored footage, viewers will be able to see chronicles with athletes who participated in the 1980 Olympics, the streets of Moscow in those years, amateur footage, press materials, and other archives.
Read materials on the topic:
LETI Created a Method for Recognizing Parkinson's Disease by a Drawn Spiral
