Neural networks of Yandex improved the quality of Soviet cartoons


The company “Yandex” has presented ten classic cartoons of the studio “Soyuzmultfilm” in good quality. The image of films is improved with the help of DeepHD technology, a service that increases the resolution of a picture without reducing sharpness, detail and quality with the help of neural networks.

When restoring old movies, the most popular task is their “coloring” or – which is much more difficult – to restore the original color. Automatic algorithms for solving this problem have been developed since the middle of the last century, and recently Disney managed to achieve in this task a better quality and realism.

Another goal is to improve the quality of old movies in such a way that they match high-resolution screens. This technology was recently launched by the developers of Yandex. Movies in DeepHD are processed using two neural networks. The first neural network removes noise and artifacts, which could remain when digitizing the old film. The second neural network – generative-competitive – creates (using the generator) improved images of individual frames in a larger size and compares them (using a discriminator) with “stretched” original frames. With the use of this technology, it was possible to increase the sharpness of the staff and improve the details, due to which the quality was noticeably improved.

According to the developers, this technology can be used to improve the quality of any other video clips: for example, by the Victory Day this year the service launched high-quality films about the Great Patriotic War. In addition, the technology can also be used to improve static still cameras (however, for this there are other services ).

Details on the DeepHD technology were also told on the Yet Another Conference 2018 (YaC-2018) – a conference that Yandex arranged at the end of May. Also on this day the company first showed Yandex.Station – a smart column equipped with voice assistant Alice.

Back to top button