Applying Objective Quality Metrics in Your Own Encoding
Watch the complete panel from Streaming Media East Connect, How to Fine-Tune Your Encoding with Objective Quality Metrics on the Streaming Media YouTube channel.
Learn more about optimizing your encoding at Streaming Media West 2020.
Read the complete transcript of this clip:
Jan Ozer: How do you use these techniques? If you're Netflix, you can use this on a per-title basis. Very few companies can do that; you just don't have that kind of money to spend on each video. I've tried this several times with category-specific encoding. What has worked is, looking at a major network that had talk shows, game shows, and then action shows, we were able to take their fixed encoding ladder down significantly--30 to 40%--for channels that use exclusively game shows and talk shows without viewers even noticing. And we proved that concept using the techniques that we just described.
Essentially, with the talk shows, when I started working with them, their target fixed rate in the fixed encoding ladder or the top fixed rate in the encoding ladder where their ceiling was eight megabits per second, for all files. And what I was able to show is that, with talk shows and most game shows, you could get to 93 VMAF points, then 95 at 5 megabits per second. Once we rolled it out and tried it, we found that customers didn't notice the difference. On the other hand, there were a lot of other shows that we couldn't get to VMAF 93 at 5 megabits per second. So we left those at 8.
We also had good success using these techniques with a training company that had some real-world videos that were actual people talking, and also videos that were PowerPoint or a screencam-based. And they came up with different ladders for both of those. That worked very effectively.
We worked with an online training company that had some videos of bicycle races, and we showed that those had to be, 1080p, very high bitrate. But some of the gym work that they were doing say for yoga or stretching or recovery, could be done at 720p and much lower bitrates, and viewers wouldn't notice the difference.
So that's where per-category worked. Per-category doesn't work if you're trying to separate out action movies or other kinds of movies. There's just too much differential within each category; separate ladders for different animations vs. movies also didn't work. maybe if you're a cartoon network and you have very specific kinds of cartoons, you can come up with different ladders. But looking at the differences between Sintel, Big Buck Bunny, and other animated movies, it's just too big to come up with a different ladders for different categories.
Unlimited file analysis and Python interface highlight the new features
Jan Ozer explains how to choose presets and tweak encoding parameters when prioritizing speed or quality, and how to find the perfect balance.
Jan Ozer explains how to use constant rate factor encoding and per-title encoding to find the optimal data rate for encoding different types of video files, ranging from talking heads to high-action movies.
Jan Ozer outlines the basics of objective quality metrics, from their recent evolution to their method for predicting how human eyes perceive video quality to the best metrics available now.