View Full Version : If CRF=18 provides enough quality...
Patrick Bateman
17th March 2009, 08:50
If CRF=18 provides enough quality that it's hard to notice any loss for most people, then what's the point of the other CQ presets if all are CRF=18?
Dust Signs
17th March 2009, 09:20
What makes you think "most people" don't see a difference? Did you carry out any tests or is it just a wild guess? Considering your source and your TV equipment alone can make a huge difference, p.e. a DVD source on a really big LCD might might not be "transparent" any more more with 18, but maybe with 16. And don't forget that's it can vary significantly from person to person, subject to personal preference.
Dust Signs
Patrick Bateman
17th March 2009, 09:35
What makes you think "most people" don't see a difference? Did you carry out any tests or is it just a wild guess? Considering your source and your TV equipment alone can make a huge difference, p.e. a DVD source on a really big LCD might might not be "transparent" any more more with 18, but maybe with 16. And don't forget that's it can vary significantly from person to person, subject to personal preference.
Dust Signs
Because many people on this forum say so.
Zwitterion
17th March 2009, 11:55
IIRC, the same CRF value is supposed to deliver the same quality across all settings. On faster settings the resulting file would be bigger than on slow settings but look nearly the same.
However, the settings are not calibrated for that behaviour (yet?) so you'll see differences although your CRF is the same, especially when changing the PsyRD and AQ settings.
J_Darnley
17th March 2009, 13:43
The different presets (from I assume MeGUI) choose different encoding options either for compatibility reasons or speed/quality reasons. The suggestion of --crf 18 by many is like lame's suggestion of -V2. It is a high (IMO) starting point. If you want more quality, lower the value, if you think the file(s) is(are) too big raise the value.
However, using x264 is far more complex because it has three psycho-visual options (AQ, psy-rd and psy-trellis) and one other quality altering setting (subme). The other options also affect the quality but to a lesser degree than those.
Dark Shikari
17th March 2009, 13:46
IIRC, the same CRF value is supposed to deliver the same quality across all settings.No it isn't. CRF is an arbitrary constant quality that is only really sure to be constant at a given set of settings. Slower encoding settings will improve quality-per-bitrate at a given CRF--but might make both quality and bitrate higher, or both quality and bitrate lower. CRF is just a ratecontrol mode without a target bitrate, not magic faerie dust.
(Is it just me, or I have I posted this same statement about 4 times this week?)
LoRd_MuldeR
17th March 2009, 15:34
@Patrick Bateman:
Even if CRF=18 looks "transparent" for most people, that doesn't mean everybody is going to use CRF=18 for his/her encodes ;)
For example CRF=18 will certainly eat significant more bitrate than CRF=22 does. Anyway, the quality delivered by CRF=22 is tolerable for many purposes.
Forcing people to use a certain CRF value would be stupid. People need to decide what value suites their needs best.
With your logic, the developers could simply hardcode x264 to CQP=0 (aka "lossless mode") and everybody would get a 100% transparent encode for sure.
Would people agree with that? Certainly No! As "lossless" is not reasonable for most purposes.
The same applies to CRF=18: It may be a good idea for "high quality" backups without size restrictions, but may be a horrible idea for anything else...
Sagekilla
17th March 2009, 17:33
Like LoRd_MuldeR said, crf 18 is considered a "baseline" in that for most people, it represents a very high quality encode. Other people may find this excessive and use crf 19, 20, etc. Or, they might find the video has bad artifacts still and decrease it still.
But, you need to decide what's best for your eyes. You can use crf 20 as a good starting point, and keep increasing it until you find the artifacts intolerable.
vBulletin® v3.8.5, Copyright ©2000-2012, Jelsoft Enterprises Ltd.