Video: How to Find the Optimal Data Rate for Your Video
Learn more about per-title encoding at Streaming Media's next event.
Watch the complete video of this presentation from Streaming Media West, T101: HOW-TO: Fine-Tuning Your Adaptive Encoding Groups with Objective Quality Metrics, in the Streaming Media Conference Video Portal.
Read the complete transcript of this clip:
Jan Ozer: Per-title encoding is the technology that creates a different encoding ladder for each piece of content or each category of content. What I'm going to walk through here is how I would do that for either specific bits of content or specific categories of content.
For this I use constant rate factor (CRF) encoding. CRF is an encoding mode available on X.264, X.265, VP8, and VP9. CRF encodes to a specific quality level and not to a data rate, so you can use it as a gauge of encoding complexity.
If you use it with CAF in a technology called CAF-CRF, you can also use it as a per-title encoding technique, and we're not going to cover that.
So what you do is you choose a range. The range that corresponds with what I'm going to call Hollywood quality--and you'll see why I say that in a second—also known as CRF23.
This is what I did for the book that I just mentioned a few minutes ago. I had eight different source files. I encoded them all to CRF23 and then I measured the VMAF score.
Here we see the data rates here range from one megabit per second (1 Mbps) for a simple tutorial, which is PowerPoint and talking head, all the way up to 6.1 Mbps for a 30-frame per second, scary movie-type video clip, and 5.168 Mbps for Cintel, and then 4.747 Mbps for Tears of Steel, These are about 24 frames per second.
So what does this tell you about the encoding complexity of the different clips? Which clip here is the hardest to compress? The one with the highest data rate.
If we look at the VMAF scores we see that the VMAP scores are all relatively similar. One is 92, another is 95. They're all in this same range, and the standard deviation is 1.39.
What that tells you is that a CRF23 encode will produce a file that gives you a VMAF score of around 93, which is our target. Because that's what we saw recently from RealNetworks.
Basically, if you use CRF23, you can predict, how to figure out the quality that will be non-objective to your viewers? What's an appropriate data rate for your 1080p highest quality file? If you use CRF23, you get that.
Related Articles
Streaming Learning Center's Jan Ozer lays out the basics of objective quality metrics for encoding assessments in this clip from his presentation at Streaming Media West 2018.
27 Feb 2019
Streaming Learning Center's Jan Ozer compares four approaches to testing video compression quality in this clip from Streaming Media East.
20 Jul 2017