The Daala Video Codec Still Needs At Least Another Year Of Development

[quaz0r]

Does anyone know at what point the theoritical maximum precieved video quality will be reached so that codec gains will begin to be used primarily for smaller file sizes rather than higher quality?

higher quality and smaller file sizes are one and the same. if one was to say the quality increased, then the same quality as before would be achieved at a smaller file size. if one was to say that file sizes have gotten smaller, then the same file size as before would provide higher quality. moral of the story: with lossy encoding, just think of it in terms of quality.

[profoundWHALE]

That looks quite poor, actually

I assume it is excessively compressed in order to keep it streamable...

This is actually a proof-of-concept and they weren't really going for performance at all. It was more of a, "Can I do it?" and the answer is yes, yes they can.

[Prescience500]

higher quality and smaller file sizes are one and the same. if one was to say the quality increased, then the same quality as before would be achieved at a smaller file size. if one was to say that file sizes have gotten smaller, then the same file size as before would provide higher quality. moral of the story: with lossy encoding, just think of it in terms of quality.

I understand that. I'm asking from a networking prespective because video is such a bandwidth hog. I figured that after, for instance, 8k, ramping up the resolution would no longer be visible to the human eye unless you were really close to the screen or the screen was friggin huge. After that, codec improvements might yield to only smaller file sizes (at least in what is primarily sent over the Internet); therefore, less pressure on the Internet. Granted, everyone who hosted lesser quality video would probably move to higher quality video before such benefits might manifest.

[profoundWHALE]

An Update on the Daala stuff



(from page 15 in these slides)

Lots of work to do
● These results are with
? No B-frames or altref equivalents
? No intra mode in our motion search
? No motion compensation blocks larger than 16x16
? No transforms larger than 32x32
? No deringing filter (pending)

It's looking good so far