Manual restoration movie and cartoons is a long and painstaking job. Fortunately, neural networks have learned to enhance images in real-time. For example, this is good technology DeepHD from “Yandex”. Its effectiveness has been proven in may 2018 — it has improved the clarity of the seven old movies about the great Patriotic war. Now neural network “Yandex” has reached the Soviet cartoons.
The neural network has processed 10 works of the Studio “Soyuzmultfilm”:
- “The scarlet flower” (1952);
- “Golden antelope” (1954);
- “The snow Queen” (1957);
- Dyuymovochka (1964);
- “The Bremen town musicians” (1969);
- “Umka looking for a friend” (1970);
- “Mowgli” (1973);
- The Nutcracker (1973);
- “Journey ant” (1983);
- “Kitten named woof” (1976-1982).
The neural network has not made any changes in shots, and just regained conceived by the artists details which are lost when you convert film to digital format. On the project page DeepHD allocated 4 technology features: real-time, increasing the resolution while preserving detail, denoise and defects, but also an equally good job with images and video.
Technology DeepHD based on generative-adversarial neural networks. In the first stage of processing of one neural network gets rid of the defects caused by compression. The second stage included other neural network — generator, increasing the size of the image while retaining sharp detail. Details about work DeepHD written in the developer blog.
In addition to the old works, DeepHD available and a new art and animated films. To find them, “Yandex” offers to enter search queries “in deephd movies” and “cartoons in deephd”. As the technology is able to operate in real time, in high quality available even TV channels.
Have you ever watch the restored movies in DeepHD? Their experiences can be described in our Telegram chat.
The neural network “Yandex” has improved the quality of Soviet cartoons
Ramis Ganiev