Slow Motion Video Using AI

Despite today's smartphones being capable of recording videos in slow motion at up to 240 frames per second, several obstacles remain. The biggest of these is the lack of memory on the device and the significant performance demand, which rapidly drains the battery.

However, Nvidia has found a solution to this problem. They have created a system based on deep learning that transforms video from a speed of 30 frames per second to 480 frames per second. This technology allows for much higher quality slow motion than traditional methods. The slow motion effect operates on the principle of two convolutional neural networks—interpolating intermediate frames between individual images.

The first neural network predicts how pixels will move from one frame to the next, thus creating 2D vectors of these movements.

The second neural network, which predicts the visibility map, eliminates those pixels that should be obscured by objects.
Yet, even though you can definitely tell the difference between video slowed down using a neural network and video shot with a professional camera in slow motion, it is a fantastic technology with promising applications for the future.

Author: Dmytro Molokoiedov

https://youtu.be/MjViy6kyiqs
Source: https://habr.com/post/414953/