Some raw numbers on SSIM, CRF bitrates etc...

[Blue_MiSfit]

Hey folks.. I've been bored recently and have done some fun little studies with x264. Here's some raw data to grok, if you're interested.

A disclaimer - I don't claim for any of this to be reproducible or terribly scientific. I just messed around with clips from these sources, about 1000 frames each. I don't even have the source clips or transcodes available. So, take all this with a large chunk of salt.

http://imgur.com/VzW38.png

http://imgur.com/wWSCc.png

Derek

[nurbs]

CRF encoding isn't much slower than CBR encoding. Encoding 7.5 Mbps is much slower than encoding 2 Mbps with everything else being the same. CABAC encoding time scales linearly (IIRC) with bitrate, other things probably too.

[Jawed]

In your first experiment, since you started at CRF 16 and worked down from there, setting qpmin to 0 or pegging it at, say, 6 less than CRF might provide interesting results. Your experiments were bumping up against the default qpmin of 10.

[Blue_MiSfit]

@nurbs: Oh yeah, why didn't I think of that? :p

@Jawed: Exactly. I consider qpmin 10 to be a very good saturation point.

Derek