Help me see the light...

[Jones_R]

I've been spending a lot of time the last two days encoding and subjectively testing the results on my 21" monitor.

I tried a lot of features, which suppose to make really noticeable differences (according to what I read here and in the guides) to the quality, like different VHQ modes (like going from 0 to 4), MPEG vs H.263, DivX 5.05 vs Xvid 24/06/03(koepi).

To tell the truth, I find it very hard to find real differences between all the output movie files. At best I can say that I can feel that there are some small nuances coming in to play, but I'm not sure that I'm not just imagining them (and I did a lot of A/B testings).

What am I doing wrong ?. Can you give me some clues for what I need to look for in order to test picture quality differences ?. Is there a special test material which will help me see differences more clearly ?. Don't get me wrong, I don't think that the results are perfect (I can spot artifacts, when I compare it to the original source), but the artifacts seem to be more or less the same no matter what I try to do. And they are not too bothersome either (at least not with my small monitor).

One area I can see differences is regarding speed. For example, encoding using VHQ 4 take me ~4X longer than when VHQ disabled (but this is already talked about). The fact that after waiting for VHQ 4 to finish (a LONG wait) I can't even tell the difference from VHQ 0, drives me nuts!.

I appreciate any help. Thanks.

PS.
My target is 3 cds, since I include the original AC3 (or Dts track, if available), so, the average bitrate I use is usally around 2000 (and I do 2pass).

[Alxemi]

2000 is a very high bitrate... Possibly the two codecs are saturated
Try a lower one to see the diferences (lower than 900)
What resolution are you using?

[Jones_R]

Alxemi, thank you for you answer.

2000 is a very high bitrate... Possibly the two codecs are saturated. Try a lower one to see the diferences (lower than 900)
But 3 cds are really not a trouble for me. Once I'll get a DVD burner, I'll be able to fit 2 movies per DVD disc.

What resolution are you using?
I always use ~720 for width, and for height, according to the aspect ratio of the movie (for example, for 2.35:1 movie, I use 304 for height).

[Soulhunter]

Hi Jones_R

Maybe you followed some other threads...?

Heres much discusion about the codec differences... 1, (http://forum.doom9.org/showthread.php?s=&postid=356071) 2, (http://forum.doom9.org/showthread.php?s=&threadid=58109&perpage=20&pagenumber=1) 3, (http://forum.doom9.org/showthread.php?s=&threadid=58626) 4, (http://forum.doom9.org/showthread.php?s=&threadid=57687) 5, (http://forum.doom9.org/showthread.php?s=&threadid=57929) 6 (http://forum.doom9.org/showthread.php?s=&threadid=39573)

Whats better... DivX, XviD, WMV9 etc.

Hmm... I could tell you what I like most, maybe you've read this in a other thread before:
I use XviD for HQ-encodec (means QB at Quantizer2... for me) DivX or XviD for Med.Quality-encodes (bitrate 2000-3000 kbps) for LQ encodes (1000-2000 kbps) I use WMV9 !!!

In my tests XviD has the highest max. possible quality...
WMV9 is superior in produceing blockfree videos with low bitrates...
For medium bitrates WMV9, DivX and XviD are close togrther... WMV9 and XviD (with MPEG matrix) have a more detailed picture than DivX... But DivX produces less blocks than XviD...
WMV9 is between this, but a little bit unuseable (no support for standalones etc.) choose what you prefeer... Sharp Image (XviD) Userfriendly (DivX) or Smooth&Blockfree (WMV9)

Have no exp. with RealV9 till now... So I cant tell much about It...

But this all is only my own prefeerence...


If you look for a good method to find out the best overall quality you can use a tool to calculate the PSNR... Use google (http://www.google.de/) or an other search engine to find the tools to do this...
Or ask other PPL here ;)

A good method to see the difference of the max. quality of a codec, is to encode some frames at Qualitybased-Mode, save a special frame of your favor as BMP from all resulting encodes, and use a Picture programm with slideshow function (like iPhotoPlus) to skip fast between 2 frames... So you can see the smallest difference !!!

I did it with a Gladiator trailer... XviD wins at Quantizer2 and also at Quantizer1... (In my tests... Do your own...)

[PS] Whats a the best tool to calculate PSNR ??? PSNR4AVI ??? Or else...?

Bye

[Jones_R]

Soulhunter, thanks!, you gave me enough to keep me busy for a while...
:)

[temporance]

@Soulhunter,

Interesting comparisons, but what codec versions and codec did you use to come to these conclusions? At the moment, divx and rv9 are making big improvements - e.g. RV9 EHQ, divx slow[est] and new pv modes.

A good method to see the difference of the max. quality of a codec, is to encode some frames at Qualitybased-Mode, save a special frame of your favor as BMP from all resulting encodes, and use a Picture programm with slideshow function (like iPhotoPlus) to skip fast between 2 frames... So you can see the smallest difference !!!To be fair, you should always consider the number of bits that are used by that frame. If one codec looks better but uses more bits, then maybe it's not such a good comparison.

For PSNR, I like to use Avisynth's compare function. Latest avisynth CVS code will generate an "overall" PSNR which is more reliable than average PSNR. Having said that, PSNR is not such a good metric to use anyway! It's best when looking for small differences and small changes. At low bitrates, or when a codec uses psychovisual tricks to make a better looking picture, PSNR is not so good. Some better metrics are JND and VQM, but nothing is perfect in this game.

Like I said though, you should always consider filesize when comparing codecs' quality. I like to plot a scatter graph in MS Excel with one data series for each codec, using different constant quality setting to make each point in the data series (e.g. QB encode with Q=2,3,4,5,6...). On the X axis, I have filesize). On the Y axis I have PSNR or whatever other metric I'm using. This sort of plot reveals a lot.

[Soulhunter]

@temporance

About my comparsion...

Small research on my HD....
Yap I saved the results, to compare them with futur comparsions ...

Here are the settings:

--------------------------------------------

XviD-Koepis-24.06.03
Settings: MPEG-Matrix/QB-Quantizer2/Default
1.43MB

DivX5.05
Settings: QB-Quantizer2/Default
1.60MB

WMV9
Settings: Q-98%/BestQuality/Default
1.70MB

XviD-Koepis-24.06.03
Settings: MPEG-Matrix/QB-Quantizer1/Default
3.62MB

DivX5.05
Settings: QB-Quantizer1/Default
3.25 MB

WMV9
Settings: Q-100%/BestQuality/Default
3.72MB

--------------------------------------------

IIRC the size based of an 250 frame clip...

Sounds short, but for fixed quantizers the lenght plays no role... ;)

The 98% WMV9 setting is maybe a little bit to much... (97% would reach the right size !?!)

About the thing with "the number of bits that are used by that frame"
In this test I tryed to see whats the MAXIMUM quality, that the codecs can reach... So as it should use the same max. quality for all frames, Its not such a big thing... (I think)
You can correct me if Im wrong !

Its also a good idear to chose more sample-frame to compare... ;)

But, sure the result will differ with a desired size in 2Pass mode...

No matter... Heres my impression:

At max. qulity (Q=100/Quantizer1) WMV9 has the biggest size, and DivX the smallest. DivX has the lowest detail and XviD the highest (followed very close by WMV9).

At Q2 (Qunatizer2/Q=98%) WMV9 is also the biggest, but XviD the smallest.
But the best image quality has still XviD, and the lowest DivX.

So for me XviD has the best max. image quality (close followed by WMV9) !!!
DivX max. quality is the lowest in this test...

PS:
[I've never tryed RV9, so I cant talk about it...]
[I think, with "improvement in DivX" you mean this new betas like Kaukura...!?!]
[Yeah, there are improvements... But I want wait till the beta phase is over ;)]

Bye

[temporance]

Testing the max quality that a codec can generate is one rare case where PSNR will work very very well! Certainly better than using your eyes to judge.

Beware that the differences are likely to be due to the various colourspace conversions, postprocessing, rendering and overlay processes that occur after decoding. The lower the quantizer, the more these processes will affect the result.

[Soulhunter]

Yeah... I played around a little bit with this PSNR stuff, and Ive also followed some other threads about this stuff...

But when this tests try to tell me different result than my eyes... Ive to say, I belive my eyes... Because I watch the movies with my "eyes", and not with a "test processing" ;)

No matter... I think both tests are very good (PSNR/EYE) :D :D :D

[PS: The "slideshow" test procedere is really complete other thing than watching the movie... Try it... I swear you will see very last lost "detail" !!!]

Bye