Nvidia’s AI Turns Normal Videos into Smooth Slow Motion Videos

Nvidia recently announced AI that can make your standard 30-fps video into a super slow motion 240 fps video.

It uses deep learning to create intermediary frames between two consecutive frames in a video. If a video is stretched out with editing software, let’s say 30 fps video into 240 fps slow motion video, it can appear choppy.


ALSO READ

Nvidia AI Can Restore Damaged Images


Nvidia’s solution avoids the choppiness using AI.  It produces mostly accurate results, but at times it messes up with its prediction (as seen in the video below).

The working principle behind this AI is that after being trained for a specific shot, its machine learning can predict the motion of an object and adds the missing frames on its own.

Manual Slomo VS Super Slomo

It compares each frame and can add up to 7 intermediary frames making the video smoother when slowed down.

Manual Slow Motion Vs Super Slow Motion

Pros

  • High-quality regular video can be changed into a high-quality slow-motion video, whereas the standard method of capturing slow-motion videos on a phone can lower down the exposure of the video.
  • You’d be able to add slow-motion effects to smartphone footage after it’s been recorded.
  • Nvidia’s AI can produce up to seven intermediary frames.
  • Can save tons of storage on your device and processing power, as it would be able to record high-quality slow motion videos for devices with limited hardware.

Cons

  • This technology is still in refinement phase as it is impossible to predict motion or behavior of real-world objects.
  • There are few limitations, for example, it being AI it needs to be trained to handle a specific type of footage in a specific type of environment.
  • It’ll only produce frames when it is given sufficient amount of data which can vary from device to device.

NVidia’s approach towards creating slow-motion videos seem more practical and hence we can expect to see this technique being adopted on our phones in the near future.

The researchers will present their work at the annual Computer Vision and Pattern Recognition (CVPR) conference in Salt Lake City, Utah this week.



Get Alerts

Follow ProPakistani to get latest news and updates.


ProPakistani Community

Join the groups below to get latest news and updates.



>