Recently, a young man in Beijing used an AI program to restore a film that recorded the daily life of Beijingers a hundred years ago.

The video material comes from a documentary film released by the People’s Daily four years ago, shot by a Canadian photographer between 1920 and 1929. After AI processing, the original figures with monotonous colors and blurred outlines in the video have been restored to have clear faces and smooth movements. With the addition of realistic sound effects in the later stage, the audience can see Beijing in 1920.

The author Dagu told the media that there are three AIs in total, the first of which is responsible for complementing frames (DAIN), the second is for resolution amplification (ESRGAN), and the third is for coloring ( DeOldify), these divisions of labor are used together, and some small free plug-ins are used in the middle.

Regarding the difference between AI repair and manual repair, Otani said that there is the same working logic behind it, and the advantage of AI repair is that it is fast. At present, the stage of artificial intelligence is still relatively elementary, and it still needs to learn in the future to become more mature. For example, the color in the video is the result of calculation and inference, and it cannot be completely and accurately matched with the facts.

There are many similar things. Last year, the treasure of human civilization, Notre Dame Cathedral, suffered a fire, and some people proposed to use AI technology to repair it. From a technical point of view, AI technology can restore and restore objects or non-objects to a certain extent through the powerful capabilities of high-performance computing platforms and AI artificial intelligence algorithms to achieve the desired effect.