Apple has quietly acquired a Mountain View-based startup, WaveOne, that was growing AI algorithms for compressing video.
Apple wouldn’t verify the sale when requested for remark. However WaveOne’s web site was shut down round January, and a number of other former workers, together with one in all WaveOne’s co-founders, now work inside Apple’s numerous machine studying teams.
In a LinkedIn publish printed a month in the past, the corporate’s former head of gross sales and enterprise improvement, Bob Stankosh, introduced the sale publicly.
“After nearly two years at WaveOne, final week we finalized the sale of the corporate to Apple,” Stankosh wrote. “We began our journey at WaveOne, realizing that machine studying/deep studying video expertise might doubtlessly change the world. Apple noticed this potential and took the chance so as to add it to their expertise portfolio.”
WaveOne was based in 2016 by Lubomir Bourdev and Oren Rippel, who got down to take the decades-old paradigm of video codecs and make them AI-powered. Previous to becoming a member of the enterprise, Bourdev was a founding member of Meta’s AI analysis division, and each he and Rippel labored on Meta’s pc imaginative and prescient crew accountable for content material moderation, visible search and feed rating on Fb.
The place it considerations commonplace algorithms for compressing and decompressing video, the compression occurs on the content material supplier’s aspect (e.g. YouTube servers) whereas end-users’ machines deal with the decompressing. It’s an efficient strategy, however new codecs require new {hardware} specifically constructed to speed up compression or decompression, making algorithmic enhancements sluggish to propogate.
Picture Credit: WaveOne
WaveOne’s principal innovation was a “content-aware” video compression and decompression algorithm that might run on the AI accelerators constructed into many telephones and an rising variety of PCs. Leveraging AI-powered scene and object detection, the startup’s expertise might basically “perceive” a video body — permitting it to, for instance, prioritize faces on the expense of different components inside a scene to avoid wasting bandwidth.
WaveOne additionally claimed that its video compression tech was sturdy to sudden disruptions in connectivity. That’s to say, it might make a “greatest guess” based mostly on no matter bits it had out there, so when bandwidth was out of the blue restricted, the video wouldn’t freeze — it’d simply present much less element for the period.
WaveOne claimed its strategy, which was hardware-agnostic, might cut back the scale of video recordsdata by as a lot as half, with higher positive factors in additional complicated scenes.
Traders noticed the potential, apparently. Previous to the Apple acquisition, WaveOne attracted $9 million from backers together with Khosla Ventures, Vela Companions, Incubate Fund, Omega Enterprise Companions and Blue Ivy.
So what may Apple need with an AI-powered video codec? Effectively, the apparent reply is extra environment friendly streaming. Even minor enhancements in video compression might save on bandwidth prices, or allow companies like Apple TV+ to ship greater resolutions and framerates relying on the kind of content material being streamed.
YouTube’s already doing this. Final 12 months, Alphabet’s DeepMind tailored a machine studying algorithm initially developed to play board video games to the issue of compressing YouTube movies, resulting in a 4% discount within the quantity of knowledge the video-sharing service must stream to customers.
Maybe we’ll see related improvements from the Apple-owned WaveOne crew quickly.