【Technology Observer】
◎Yang Xue
Recently, the 14th Beijing International Film Festival opened, and the new 4k versions of the movies "a plan " and " " premiered in the "Tribute·Repair" section. The familiar picture reminds many people of the warm time spent watching movies in front of the TV. The difference is that this time the audience can fully see the "lightning-fast" martial arts movements.
In the movie "Plan A", there is a fight scene: Facing Sammo Hung holding a sword high, Dewey, who was already lying on the ground, kicked the knife away and turned defeat into victory. In fact, this scene does not conform to common sense. There is a key link missing in the middle - how did Dewei stand up and kick away the big knife in an instant?
Now, intelligent frame insertion technology has "uncovered" the mystery: between the two martial arts movements, the middle value of the action predicted by AI is inserted - Dewei puts his hands on the ground and prepares to stand handstand, then exerts force on his waist and uses his feet to twist Sammo Hung's hand, Sammo Hung lost strength and the sword fell.
It is reported that the China Film Archive, Douyin, and Volcano Engine have jointly restored 100 classic Hong Kong movies, and the above two movies are among them.
The development of AI technology, especially the widespread application of AIGC technology (using artificial intelligence technology to generate content), has given old movies a "new life". The powerful generation ability and rich prior knowledge of large visual models can not only greatly improve the repair efficiency and eliminate defects in the original film, but also comprehensively improve the clarity, color, smoothness, etc. of the film.
Film restoration is a trivial job with complicated processes. Even for an experienced film restorer, it still takes a month to restore a film.
Nowadays, there is a "division of labor" in the film restoration process. During this restoration work, the Volcano Engine technical team communicated and discussed with the restorers of the Film Archive. During the digital restoration process, the engineers used AI marking technology to reduce the duplication of work by the restorers.
Old movies carry the youth memories of a generation of viewers. Thanks to the continuous improvement of restoration technology, precious image data can be better preserved. Those classic scenes from the old days have been brought back into the public eye, allowing young people to have the opportunity to experience the charm of old movies, and using technological innovation to help cultural inheritance.
It is worth mentioning that while solving various practical problems in the restoration of old movies, AI technology itself is also constantly evolving. After the large AI model learns massive data, it can extract a large amount of knowledge from it and "store" it in the model parameters. The generation space is richer and it can handle more repair scenarios in the future.
(Science and Technology Daily)