I contacted CR with your assumption months ago but here you are again... so here this is again. Its easy to go to lnu and ask yourself if you actually want answers.
so here it is from jan 2021 below.
jan/2021
I asked Chris Richardson/jodi for a detailed response on this for you bcause you seem to have an aversion to going to the source to answer questions. Heres CRs response again below. I hope that helps provide more context for you.
***************
It is true that Linius primarily outputs HLS these days. When the company first started, we used a somewhat proprietary MP4 format (it was actually to MP4 standard, but few commercial players supported the standard settings required to achieve what Linius does). HLS (and to a lesser extend DASH) is just the standard way video is played today. By outputting in that format, we ensure that no client-side changes are required to play our virtual videos — there’s native support for things like trick-play and DRM. The Linius magic is on the back end, and in the assembly of these outputs.First of all, to “just” take existing video and assemble it into useful new videos is a non-trivial process; it’s not something a knowledgeable programmer (or even a team of them) could just “hack together” — if you don’t believe me, look at what happened when IBM tried to do just that for tennis and golf. Second, what happens when your original video files are uncooperative? The standard for recorded HLS is ten seconds per segment. Ten seconds doesn’t sound like a lot, but what happens when the clip you’re seeking starts 8 seconds into a ten second clip? Even if someone could hack together a solution to assemble that, eight seconds of nonsense is a terrible user experience. Again, don’t believe me? Try going to a web page that takes ten seconds to load and tell me what you think? Worse, what if that same thing happens to the second, and third, and nth segments that you’re trying to assemble? All of a sudden you a video with eight-second gaps of nonsense at every transition. Linius’ solution enables you to start exactly when you want. Finally, what if what you’re after isn’t HLS encoded already? Or what if you’re assembling bits from different encodings? Part HLS, part DASH, part raw MP4? And what if there are different codecs in use? VP9 in one segment, h.264 in another, etc….?There’s a ton of intellectual property in the Linius search and assembly functionality. The assembly side is well covered by patents, but even on the search side this is non trivial. Yes, Google, and Amazon, and Microsoft have video-recognition algorithms, and they’re getting better and better every day. But even as they get better, they’re still … not particularly good. And worse, they’re not particularly good in different ways. Although only parts of this are likely patentable (and we’re seeking patent protection for them), our ability to take data from multiple different sources (e.g., from Microsoft for face recognition, from Amazon for object detection, from Google for speech-to-text, from a sports-data provider for event highlights, etc… etc…), aggregate it, and make it searchable in a meaningful way represents years of IP development that is certainly at a minimum “trade secret” level.We have yet to see anyone else who can do what Linius does; to duplicate it would, as an understatement, be a non-trivial exercise; and, until someone comes up with something new, would almost certainly be a violation of our existing patents.
Expand
(20min delay)
