View Full Version : CoDecs - Best, i mean BEST Quality
CQ
12th September 2004, 21:57
hei. i`m running a codecs test for a pc magazine (yeah, PC Magazine Ro) and i wanted to know wich are the best settings for these codecs. Speed isn't important, quality... yes! So do youre best. gmc, slow, adaptive multiple, dropping the frames, 2 pass encoding bla-bla... everyone, wich settings are the best with On2 VP6, ffdshow's DivX3, Wmv9, H.264 (the mpegable's). If you think that other codec worths to be in... tell me.
ps. excuse my english but my second language is french and english comes on the last place :)
please, answer. any answer (good or wrong is welcome - i will sort them :)
Manao
12th September 2004, 22:14
Sorry to sound harsch, but you should have read the forum's rules (http://forum.doom9.org/forum-rules.htm) (#12 especially : "do not ask what is best" ). For the rest, a good start point is to read the Doom9's guides ( you can find it in french here (http://french.doom9.net/) ), and to have a look on the settings used by Doom9 to make it own test (http://www.doom9.org/codecs-203-1.htm).
Now, on your codec choice, you have at least to add XviD to your list.
Edit : I read your other posts, you were already planing to test XviD, so OK.
CQ
12th September 2004, 22:22
i know you garçon :) but wich settings not wich codec. the slowest and the highest quality settings for those codecs. comprende amigo? (3 languanges in one thread, cool huh :)
Sagittaire
12th September 2004, 23:07
IMHO best setting for each codec:
1) XviD in 2 pass encoding
- Unrestricted profil + HVS Best matrix with PP0 or H263 with PP4 + Adaptative Quantisation + GMC + QPel + BFrame 2/1.50/1.00
- Ultra High + VHQ4 + Chroma Motion + Trelli + Quant interval [2-31] for all type frame
2) DivX 5.2 in 3 pass
slow "-pq 64" + H263 Matrix + Multibframe + GMC + Psy fast
3) VP62 Sharpeness profil in 2 pass encoding
use defaut setting with best quality for second pass
Use PP4 for post-process
4) WMV9 VCM in 2 pass
Maximum Quality + Complex
Use PP1 for Post-Process
5) RV10 in 2 pass
EHQ 100 + New Rate Control "Elysean"
Use HFE1 for Post-process
CQ
12th September 2004, 23:20
thanks man. do you know how much clips i'll use? about 5-8. with 10 bitrates: 250,500, 750...2500... its a big test :) and with about 50 subject (from 10 to 60 years old subjects). if you can help me any more with the mats test (ssim, psnr...) i'll apreciate that.
JohnV
13th September 2004, 00:09
What is your deadline?
Testing a mediocre h.264 implementation like Mpegable gives masses a wrong impression on what h.264 is really capable of. Masses don't understand that there can be huge differences between implementations. And that of course benefits Microsoft, VP6, Real.
I'd suggest you wait for the Ateme/NeroDigital h.264 launch or ask permission to test for example Ateme beta4 encoder. Testing the current best implementation probably helps even the other not yet so good h.264 implementations and h.264's common image in general (instead of putting it down with Mpegable).
Another point is that if you don't test Ateme/NeroDigital h.264, your results concerning h.264 quality will be immediately out-of-date when ND h.264 is released (in maybe 4-6 weeks).
CQ
13th September 2004, 01:20
JohnV, thats a good ideea. We'll find the best H.264 implementation (maybe we can gain acces to a DivX H.264 or the alpha DivX Q :) and then...
Deadline... end of oct. beginning of nov. (i'm thinking of puting it on the dec. edition becase i want to make a test that... A BIG AND COMPLETE ONE!)
And now my intentions.
This is what I plan.
1. XviD in 2Pass Encoding
Unrestricted profile, H.236 Matrix for the 250-1250 bitrates (corect me if i'm wrong with these) and MPEG for 1500-2500 bitrates (or you're HVS? whats the diference between MPEG and HVS? Its optimized for medium bitrates and high sharpness?), Advanced, QPel, GMC, Bframes... why shoundn't i use 3? I repeat, compatibility isn't a problem here. MSP - ultra high, VHQ - wide search, Chroma motion, Trellis and 2-31 for B-P and 1-31 for keyframes maybe; PP... where is this :)
2. DivX 5.2.1 in MultiPass 3Pass Encoding
Slow, Slow PSY (Alastair says that fast can be, most of times the most recomended setting), GMC, Multiple Bframes, H.263
3. WM9 - 2Pass VBR (what peak does it mean? its about streaming and buffers, right?) Complex, Perf. Better Quality PP... like XviD's... where is it? :) :) :) :)
4. RealVideo 10 "Elysian" from the RealProducer Plus 10... isn't this the best tool for spending wise every bit, right? Its more for streaming files, huh? So RMFactory (why shouldn't i use the helix producer with CL directly?) with EHQ 100... here i need more informations, i didn't use too much RV.
5. VP6... maybe a litlle tweaking? with droping the frames (doesn't matter that PSNR and SSIM are messed up its both obj. and subj. - more like 25-75% - because most people need the best visual perceptibility codec rather than watching high points in SSIM, VQM and PSNR and when it comes to watchin' it...)
Or i could do a test like this: first optimized for PSNR, SSIM...tralala and with subjective testing and then optimized for visual qua... with obj. testing too. And then, the reader could make a ideea about these. And after that he could say: Yeah, Im interested in visual quality, second test results are the ones for me.
H.264 - should I use it in my test? Yeah, It is MPEG4 Part10 but the implementations are... litlle babies! So why bothering?
DivX SBC... here, i would need a litlle (a little more) infos. And... surprise! I'll use CCE's MPEG1 and MPEG2 for comparision (the very newbies in this subject could see what MPEG1,2,4 means). Its a bad idee, right? :)
The bitrates that I'll use are: 250 Kb/s, 500 Kb/s, 750 Kb/s, 1000 Kb/s, 1250 Kb/s, 1500 Kb/s, 1750 Kb/s, 2000 Kb/s, 2250 Kb/s, 2500 Kb/s :) :) :).
There will be about 7-10 clips with very different actions (nature videos, action films, DV amateur footage - yeah, i'll get some shaky videos :)...).
The first 2 bitrates (250 and 500) are for CIF resolution and the 750-2500 for full D1.
In Sagi's test
Sagittaire, what XviD vs. VP6 or vs. DivX 5.2.1... means? Why do you compare XviD with the other codecs? And from youre test... how do you obtain the graphs (i know, in excel but the first ones: Frames Size Xvid Quant2...).
In this test, I'll write about an page for each codec (a very short explanation :( ) so if you (all of you boys and girls) would like to help me a litlle more (maybe doing a list with the good and the bad things for each one, in youre vision) I`ll be verry thankfull (I repeat from this magazine www.MyC.ro I get no money, its a pleasure so... i ain't using you for my personal income increase :) ).
Thank you verry much [especialy to the person that made this possible Doom9 - cheers mate :)] and all the persons out here.
You make it worthable (its this how its writing? :)
THANK YOU, ALL OF YOU!
*.mp4 guy
13th September 2004, 03:03
Obviously im a little new here, but im reasonably experienced with some of the codecs on the list. so ill try to give you some info.
Ok first off you should probably use hvs-best for all bitrates up to ~1750kbps after that you should definately use sixofnines max=24 matrix for the xvid tests. That is unless its a metric test in which case the default metrices are almost always best.
Second in my opinion you dont need to test divx 5.2.1 since it isnt as good as any of the other codecs, but it is popular so you might want to test it anyway.
Next dont test h264 unless you can get the nero digital beta to test with as it is far better then the rest, and because even in its current stage it has a chance to beat all the listed codecs in quality.
And last are my recomended xvid settings:
1: profile@asp lvl 5
2:mpeg custum matrice= hvs-best/sixofnine
3:adaptive quant=on
4:quarter pel=on
5:GMC=on
6:bvops=on
7:max consecutive bvops=2
8: quantizer rati=1.5
9:quantizer offset=1
10:closed gov=on
11:chroma optimizer enabled=on
12:msp=6
13:vhq=4
14:max i frame interval=10*frame rate
15:quantizers=2-31 (including i frames)
16:trellis quant=on
17:for low bitrates use uv deblocking and deringing and y deblocking, for high bitrates us only uv deblocking
18: always use two pass mode
hehe... had a rogue smilie:D
CQ
13th September 2004, 09:21
keep continuing with the coments boys!
Teegedeck
13th September 2004, 09:23
Originally posted by CQ
1. XviD in 2Pass Encoding
Unrestricted profile, H.236 Matrix for the 250-1250 bitrates (corect me if i'm wrong with these) and MPEG for 1500-2500 bitrates (or you're HVS? whats the diference between MPEG and HVS? Its optimized for medium bitrates and high sharpness?), Advanced, QPel, GMC, Bframes... why shoundn't i use 3? I repeat, compatibility isn't a problem here. MSP - ultra high, VHQ - wide search, Chroma motion, Trellis and 2-31 for B-P and 1-31 for keyframes maybe; PP... where is this :)
The HVS-matrix is designed to give the viewer the impression that the picture holds more detail than there actually is. While I used to favour the HVS-best matrix, I would nowadays only use it for actually lowly-detailed material. MPEG is always a good choice. But I'd go for Didée's SixOfNine matrix for the highest bitrates and his SixOfNine-HVS (which unfortunately has become known also as "max=24)" for the rest. But that's just me. The first valid test of custom matrices is just about in its early stages of preparation right now, courtesy of Soulhunter.
If you plan on using high-bitrate matrices like SixOfNine, prepare to have a more complicated setup. (First pass must be done at quant=3 via zone setup; using fixed quantizers in zones prevents XviD from automatically disabling stuff like trellis for speed-up in first pass, which you now have to do manually...)
C0mPr355
13th September 2004, 16:20
for VP62 use basically what sagitairre stated. However, I will add a little bit...depending on material.
under general:
mode:two-pass first pass ---> two pass best quality
end usage: VBR aka Local Playback
material: progressive
noise reduction: set to 3 if the source is very noisy...otherwise keep at default.
sharpness: I tend to favor a sharpness of 7 during encodes.
auto-keyframe: checked
max frames between keys: 360
under advanced keep everything as is, but just make sure that temporal and spatial resampling are NOT checked as they will greatly reduce the quality of both playback(temporal) and image(spatial).
And make sure that the min and max quantizer are set to 4 and 56.
bobololo
13th September 2004, 17:24
Originally posted by CQ
hei. i`m running a codecs test for a pc magazine (yeah, PC Magazine Ro) and i wanted to know wich are the best settings for these codecs. Speed isn't important, quality... yes! So do youre best. gmc, slow, adaptive multiple, dropping the frames, 2 pass encoding bla-bla... everyone, wich settings are the best with On2 VP6, ffdshow's DivX3, Wmv9, H.264 (the mpegable's). If you think that other codec worths to be in... tell me.
You didn't mention NeroDigital MPEG-4 ASP Codec in your list. Even the hype is actually focused on the new AVC / H.264 codec, the MPEG-4 ASP encoder isn't so ridiculous compared to its rivals.
superdump
14th September 2004, 00:40
Originally posted by bobololo
You didn't mention NeroDigital MPEG-4 ASP Codec in your list. Even the hype is actually focused on the new AVC / H.264 codec, the MPEG-4 ASP encoder isn't so ridiculous compared to its rivals. It certainly isn't ridiculous. It's a very good codec and its only ridiculous attribute is its immense speed. I would recommend you include it in the test.
Atamido
17th September 2004, 23:04
Bah, BEST codec for quality? Huffyuv baby!
Sirber
17th September 2004, 23:08
About RV10
You can use CL if you want if you feel comfortable with XML editing. You can also try my GUI, which is set to 65/100 by default with Elysian features.
ChaosBlizzard
18th September 2004, 21:59
Originally posted by Pamel
Bah, BEST codec for quality? Huffyuv baby!
I prefer ffdshow myself. Uses very little cpu and gives results that are better than the Divx codec. :)
Valeron
26th September 2004, 14:45
Waiting for such a great codec comparison for a long time.
Include XviD and RV10,please.
One clip test can proove little.
Personally, there have been a certain amount of good codec out there,why still insist to include a "baby" H.264 implementation but no XviD?
According to this logic,you can wait for 10 years for serveral times superior codecs compare to H.264.
Be care for your XviD setting,the output result will be vary depending on your source and you'll need the proffesional guides here.
My request:Include both WMV9 VCM and WME9.I always wonder the differences between them.
Forget WMV9 advanced profile now.It's not ready for 2 pass encoding.
My question:Will you post the result on web site?
Probably I can't buy your magazine in China.
Hopefully,the codecs you should include is completely free for personal users.:D
ac-chan123
26th September 2004, 16:17
I get the best Picture Quallity with Raw/uncompressed RGB, but the file is a little to big. Very small files i get with /dev/null, but the quality isn´t the best.
loni_blues
3rd October 2004, 15:41
@teegedeck,
If you plan on using high-bitrate matrices like SixOfNine, prepare to have a more complicated setup. (First pass must be done at quant=3 via zone setup; using fixed quantizers in zones prevents XviD from automatically disabling stuff like trellis for speed-up in first pass, which you now have to do manually...)
But surely this procedure does not affect the quality of the first pass, just its speed. Anyway, if I may ask, how can you disable trellis and other stuff manually?
Thanks!
Didée
3rd October 2004, 19:06
loni_blues:
I gave an explanation here (http://forum.doom9.org/showthread.php?s=&postid=550379#post550379), why 1st @ q3 is better for high BR matrices.
loni_blues
3rd October 2004, 21:00
Thanks Didée,
I'd already read your valuable explanation. I was just wondering how Teegedeck managed to do a fast firstpass at q3.
Thanks for your matrices, by the way!
vBulletin® v3.8.11, Copyright ©2000-2026, vBulletin Solutions Inc.