The neural network can distinguish between people and objects in the video and separately speed up or slow down their movements, while ensuring smooth interaction with each other, New Scientist reports. This can be used in advertisements, instructional videos, or in cinematography to dramatize or reduce emphasis on certain movements or events in a video.
Editing individual elements in a video requires frame-by-frame changes. This is usually done manually, like in animation, or using special tools that have limited functionality and can only perform certain programmed actions. A development team at Google and Oxford University was able to split each video frame into separate layers and taught AI to recognize people or other moving objects in them.
The developed neural network offers to select an object to which it is necessary to apply deceleration or acceleration. She then tracks objects at each level, focusing on their movements. As a result, everything in the video will happen the same way as before, and one object will start moving at a different speed. The made change does not affect the smoothness of playback and does not spoil the overall picture. It is very difficult to guess about the changes made with the help of AI.
For an example of how the algorithm works, experts created a video in which three children randomly jump into the pool. The algorithm “corrected” the video, making the kids’ jump almost synchronous.
The developers note that similar tools will appear in the future in many video editors and will allow controlling the viewer’s attention to a new level.
The widespread use of such algorithms casts doubt on the reliability of video recordings from car recorders, photo finish, video assistance to judges of sports competitions and any other video documentary materials.