View Full Version : x264 efficiency
Terka
25th June 2010, 13:42
have 7seconds DV clip.
encoding settings are:
--preset slow --level 4.1 --crf 22 --tune film
1. using TGMC 4500kb
2. making 1st frame only (same duration as 1) 326kb
3. made 1st frame moving up/dn left/right same as original clip. (via depan) 1944kb (small black borders appear because of the shift)
is 1944kb good value? for only shifting a picture?
which settings to tune to get it lower/if this is possible?
audyovydeo
25th June 2010, 14:51
are you talking bitrates or file sizes ?
cheers
a/v
Terka
25th June 2010, 14:51
filesizes. sorry.
AlekseiV
25th June 2010, 15:07
--preset slower
audyovydeo
25th June 2010, 15:55
filesizes. sorry.
If you're aiming for a file size, you should not use crf ratecontrol.
But you're trying to compare x264's efficiency vs TMPEnc.
I don't see how that can be done with file sizes.
You can compare quality / bitrate (and speed I guess, altough I don't see the point).
cheers
audyovydeo
Terka
25th June 2010, 19:19
im saying, that one, exactly the same frame, moving up/dn left/right needs a lot of bitrate.
foxyshadis
25th June 2010, 19:47
You'll always get higher efficiency no matter what codec you use if you use depan's border mirror option. Black borders kill efficiency.
Terka
28th June 2010, 10:57
yes, i know.
for me it looks like a waste of bitrate.
im saying, that one, exactly the same frame, moving up/dn left/right needs a lot of bitrate.
Dark Shikari, is there something like global motion search in x264?
*.mp4 guy
28th June 2010, 11:30
GMC gives very little gain for lots of complexity. Therefore no one but the Xvid devs properly implemented it in ASP, so no similar features are in AVC (or if they are, everyone is ignoring them again).
Also AVC's superior motion vector prediction should do just about everything GMC did, while being faster and more useful in areas GMC never helped with. GMC basically acted as an overly complicated mean removal for motion vectors. Warp points made things like rotating cameras and zooms theoretically cheaper, but never actually paid off on any real world footage. By predicting motion vectors using a fairly simple median based method (iirc), and coding the residual losslessly, you get pretty much the same performance.
TL;DR - GMC is bad, and AVC doesn't need it.
Terka
28th June 2010, 11:34
*.mp4 guy, Thank you. Do you have any recommendation for hand-held camcorder settings, or is this not necessary/no gain.
*.mp4 guy
28th June 2010, 11:55
I recommend setting x264 to use a an avisynth script as input. beyond that, there isn't that much to set in x264. There is much greater variance in camcorder footage quality then there used to be, so all I can really advise is fixing whatever is wrong with it, and denoising and stabilizing it until you can hit your target filesize/bitrate at a reasonable crf.
In my experience, you can usually compress lossless or near lossless source footage a lot more in terms of quantization level before it looks bad. So don't be afraid of higher crf factors if that's what you have to do.
Terka
28th June 2010, 13:29
So if i understand correctly, so you suggest me to use higer the crf? ill do some testing. Thank you!
vBulletin® v3.8.11, Copyright ©2000-2026, vBulletin Solutions Inc.