Nvidia: Convert a standard video into slow motion with AI
When it comes to slow motion video, everything is played at the number of frames per second. If not enough frames are recorded, the slower the video gets, the more choppy and tiring the eye will feel. Unless artificial intelligence is used to create extra frames, making the whole experience smoother.
New research from Nvidia does just that. Using artificial intelligence, a video shot at 30 frames per second ends up having 240 frames. Essentially, the AI compares two frames and creates new intermediate frames based on the movement of objects from one to the other. The way it creates the new footage isn't exactly the way the human brain would, but it produces fairly accurate (though not perfect) results.
The process will need to be optimized until it can be used commercially, but once that happens it could be used to add slow motion effects to mobile device videos after they have been shot.
“While it is possible to record video at 240 frames per second on today's mobile phones, the option is not practical, as it requires a lot of storage space and consumes a lot of battery. For this and other reasons, creating slow motion video from existing material is of great interest.”
However, there are limitations. Initially, the researchers involved in this technology went through a process of training the AI to handle specific types of video. For example, if they wanted to create a slow-motion video of a moving car, they first had to train the program with similar footage. It is worth noting that the program cannot produce an infinite number of frames. Without providing data (hence the training process), it is impossible for a computer to predict the movement of physical objects (like a car). Finally, Nvidia's system can create up to 7 intermediate frames, which is more than enough for a decent slow-motion video.
[su_youtube url=”https://www.youtube.com/watch?v=MjViy6kyiqs” width=”640″ height=”380″]https://www.youtube.com/watch?v=B7o0qA4L4So[/su_youtube]
