Filming slow-motion video requires capturing greater than the usual 30 or 60 frames per second. You may not all the time have round that may try this, however Nvidia has a workaround. It has developed a brand new AI algorithm that may take your common video and construct new frames out of nothing to create high-speed video after you’ve filmed an occasion. Nvidia achieved this feat with an array of GPUs and a neural community.
Excessive-speed video is extra accessible than ever. You don’t want multi-thousand-dollar digicam rigs to seize one thing in slow-motion when telephones just like the Galaxy S9 can movie in 240 frames per second. Nonetheless, restricted space for storing and processing energy imply you’ll be able to solely take just a few seconds of high-speed video. Few customers are going to purchase a tool designed particularly to seize high-speed video. So, Nvidia’s system can take your common video and make it high-speed by trying on the current frames. It then creates new middleman frames (as much as seven of them) so the movement is smoother while you sluggish it down.
The brand new frames are a superb approximation of actuality, however they’re not actual — a lot of the frames are laptop generated. Nvidia used Tesla V100 GPUs and the cuDNN neural community framework to construct a mannequin for processing video into high-speed video. The staff educated the community on greater than 11,000 movies of on a regular basis occasions filmed at 240 frames per second. The community discovered the right way to emulate a 240fps video to it might predict the additional frames in movies that solely have 30 frames per second.
Nvidia efficiently slowed down 30fps video of a automobile skidding throughout a street, individuals taking part in sports activities, and extra. With out the interpolated frames from the neural community, slowing down that video simply makes it appear to be a uneven cease movement animation. The community even works on video that was already filmed in slow-motion. Nvidia used footage from the YouTube collection The Sluggish Mo Guys, which is shot at 240fps already. The staff slowed these movies down by an element of 4 with the identical easy movement.
The examples proven by Nvidia are spectacular. Should you didn’t know the footage was largely laptop generated, you may not suspect something was amiss. Nonetheless, a notable limitation is the neural community can’t deal with simply any outdated video. It was educated on sure varieties of movies, so it could solely decelerate comparable 30fps movies. For instance, to decelerate video of the automobile skidding, Nvidia wanted to coach the community with high-speed video of automobiles skidding. That is nonetheless fairly spectacular for a analysis mission. Hopefully, it turns into an actual shopper expertise sometime.