View Full Version : any better settings?
uray
12th November 2005, 10:36
I have encoded movie, with these settings (cli)
x264.exe --progress --pass 1 --bitrate 390 --stats "x264.stats" --subme 6/7 --trellis 2 --ref 8 --threads 1 --filter 0:0 --keyint 350 --min-keyint 30 --scenecut 40 --qpmin 12 --qpmax 48 --qpstep 6 --directtemporal --me umh --merange 32 --sar 1294:1000 --bframes 4 --weightb --b-pyramid --b-bias 0 --ipratio 1.20 --pbratio 1.25 --qcomp 0.65 --mixed-refs --analyse all --8x8dct --progress -o
x264.exe --progress --pass 2 --bitrate 390 --stats "x264.stats" --subme 6/7 --trellis 2 --ref 8 --threads 1 --filter 0:0 --keyint 350 --min-keyint 30 --scenecut 40 --qpmin 12 --qpmax 48 --qpstep 6 --directtemporal --me umh --merange 32 --sar 1294:1000 --bframes 4 --weightb --b-pyramid --b-bias 0 --ipratio 1.20 --pbratio 1.25 --qcomp 0.65 --mixed-refs --analyse all --8x8dct --progress -o
what I want to ask is, is there any better settings to get better result with restriction to file size?..., or if i make some mistake on my settings above that is unneeded settings or settings that will lower the quality,please inform me...
these are the result of that settings, with original movie for reference
Original :
DVD 720x400 Duration ~1h.50m 4.1GB (video)
http://students.math.itb.ac.id/~uray/DVD.png
encoded movie :
544x368 anamorphic to 704x368 , size 315MB (video), same duration, container MP4, x264 ref.365
http://students.math.itb.ac.id/~uray/MP4.png
thank you
hpn
12th November 2005, 12:45
No need to use every possible option in your command line. Just skip most of them and x264.exe will use their default values.
If only the ultimate quality matters to you and you have a very fast computer and 96 hours you could try the extreme set:
x264.exe -p 3 -B 390 -r 16 -b 3 --b-rdo --b-pyramid -A all -w --me esa -m 7 --mixed-refs -8 -t 2 --progress -o insane.mp4 b.avs
leave all other options at default.
If you can't wait that long you could use --me umh instead of --me esa, -r 8 instead of -r 16, -m 6 instead of -m 7, etc... --merange 32 is in most cases worse than the default 16, so no need to use it. Also -f 1:1 sometimes provides better PSRN (quality) than the default -f 0:0, but only tests can prove it in your case. If you want 3 passes use
x264.exe -p 1 -B 390 -r 16 -b 3 --b-rdo --b-pyramid -A all -w --me esa -m 7 --mixed-refs -8 -t 2 --progress -o NUL b.avs
x264.exe -p 3 -B 390 -r 16 -b 3 --b-rdo --b-pyramid -A all -w --me esa -m 7 --mixed-refs -8 -t 2 --progress -o NUL b.avs
x264.exe -p 3 -B 390 -r 16 -b 3 --b-rdo --b-pyramid -A all -w --me esa -m 7 --mixed-refs -8 -t 2 --progress -o insane.mp4 b.avs
Note that the second line is -p 3 not -p 2, so it's not a typo.
For more tips use search.
nm
12th November 2005, 13:33
@uray:
The darker colors in your encoded video seem to be completely cut off. For example the background is dark gray in the source, but in the encoded screenshot it is black. There seems to be some bad filtering going on. Turn up the brightness and lower the contrast in your display if you can't see it.
uray
12th November 2005, 15:26
@nm :
yes that is... it was avisynth filter, i think it will improve compressibility when using trellis... is that so? and also i use convolution filter to reduce texture complexity, my target was to encode movie with result 256MB of size (including sound,duration ~1h40m) with acceptable quality, any tips for extreme low bitrate encoding?
@hpn:
- what is *"-rdo" ? i don't see it on parameter list of x264?
- i don't want to use -esa, -umh is enough, but i want to optimize another options
- are you sure -merange 16 is better than -merange 32
thanks for the replies
foxyshadis
12th November 2005, 15:48
I noticed the overdone levels too; I don't mind the background getting pushed down, but all definition on her hair is pretty much gone. Lower the black-point unless you're going for a much darker look.
Also if you want a visual comparison for your video, it'd be best to post an actual encoded video clip (and perhaps source as well), stills really don't mean much.
uray
12th November 2005, 16:07
I noticed the overdone levels too; I don't mind the background getting pushed down, but all definition on her hair is pretty much gone. Lower the black-point unless you're going for a much darker look.
Also if you want a visual comparison for your video, it'd be best to post an actual encoded video clip (and perhaps source as well), stills really don't mean much.
yes i want to wash out darker and higher complexity texture, but her hair is the side effects (what should I do?), it is ok if it acceptable as long I'll get more compressibility
yes it is original movie (without filtering) and the encoded movie is filtered before encoded not a post processing result, i just want test acceptability of movie in sense of HVS not comparison in sense of codec performance......... ups... sorry do you mean to post movie clip not a screenshot of movie clip?
hpn
13th November 2005, 00:43
- what is *"-rdo" ? i don't see it on parameter list of x264?
thanks for the replies
My bad, use only --b-rdo. RDO should be triggered when you use --subme 6/7 (-m 6/7). It is supposed to mean Rate Distortion Optimization. These are still not officially in the SVN and are only available in the Sharktooth's patched builds.
- are you sure -merange 16 is better than -merange 32
It's been reported a few times and I've also tested 16 vs. 32 and seems it's true, but just make 2 short tests with your source with both 16 and 32 and compare the global PSNR values. The bigger one is the better choice for you. A test with 2000 frames should be enough (use --frames 2000 directly in the x264 command line or temporarily add trim(40000,42000) or something to the avs script).
Alternatively you could use MeGUI plus the Sharktooth's Custom Video Profiles (http://forum.doom9.org/showthread.php?t=101813) - the HQ-Insane profile will give you almost max quality. MeGUI uses a "Turbo" option by default (you could uncheck it manually) that forces "--analyse none" during the first pass (or during 1st and 2nd in case of 3pass encode), so it's faster this way, but I haven't tested yet by comparing PSRNs if the amount of quality loss (maybe very small) makes "Turbo" a good idea. I'll test it soon when I find time :)
vBulletin® v3.8.11, Copyright ©2000-2026, vBulletin Solutions Inc.