Streaming Learning Center’s Jan Ozer makes the case for custom, per-title encoding on the Netflix model, focusing on resolution issues in part 1 of this 2-part series from his Streaming Media West presentation.
Following the model established by Netflix in December 2015, Jan Ozer argues against content providers applying a standard encoding ladder to all of their content because of the inefficiencies and waste that approach builds into the process. In Part 2 of this 2-part series, he outlines his process for handling the resolution side of the equation from mobile on up; in Part 1 he dealt with resolution.
Read the complete transcript of Ozer’s remarks in this clip:
Jan Ozer: Once we’ve chosen our data rate, what’s the best resolution at each data rate? What Netflix does is they encode it all; they encode every file at a bunch of different resolutions that are a bunch of different data rates. They started out using PSNR, they’ve since moved to the VMAF algorithm. Then they look at what’s the highest of all the files that we encoded at 3.4 megabits per second, what’s the highest quality? Our 4.6 megabit per second file will be a 1080p, our 3.1 will be 1080p. Our 19 will be 720p, 720p of the 16. The 1 megabit per second file will be 540p, 500 kilobit per second is still 540. I’d probably move that to 360 just to have a file at that resolution. Then our lowest resolution file would be a 270, not a 180. That’s how they build their encoding ladder by using objective quality metrics.
Then this is a cool graph. This shows you at any point on the data rate scale you can see what the highest quality resolution is. If we say we want a 2.5 megabit per second file, the 1080p stream is the highest quality stream. You can also see which streams are never the highest quality. A lot of people have really low resolution streams in their encoding ladder. What this tells you is these streams down here are never the highest quality, so you might as well just stop producing them. You’re better off sending a 360p stream than a 270 or 180 stream at every data rate. Then you can also see how much quality goes up, if you increase the data rate. Here, increases in data rate deliver a lot of additional quality. Here, the quality curve is starting to flatten out which means less benefit for the higher data rate.
What happens when we move to HEVC, a higher-quality codec? You get even more layers down here that never deliver the highest possible quality. If you’re moving the HEVC, you really don’t need to include even your 640×360 stream because 540 at all data rates will deliver a higher-quality image than 640×360.