NON-representatitive and probabely blunderingly made codec comparisson

[kilg0r3]

hi there,

i made a little NONREPRESENTATIVE comparisson between xvid and divx. for this prupose i encoded a full length feature film. the data is given below.


'The Curse of the Jade Scorpion (Woody Allen)'

the film contains a lot of still scenes but also some with a fair
amount of motion, especially the office scene where everybody is
rushing around.

length: 97min, res: 576x320,
video filesize/bitrate: 629926kbyte/ca.885kbit per sec

method: Smartripper, DVD2AVI, AviSynth, VirtualDub 1.4.10 (two-pass)


XviD Koepi-22052002-1
---------------
global: msp:6 - q-type:h.263 - lumi:yes (both passes)
quantization: minQ I:2 - maxQ I:16 - minQ P:2 - maxQ P:18
two pass: all at default - hinted me: no
altcc: medium - high:200 - low:45 - strength:50

DIVX Pro 5.02
---------------
gmc, b-frames, psych:strong, maxQ:12, minQ:2

to my eyes and surprise, the divx codec seems to provide more details and less artifacts with these settings. only in dark scenes, lumi seems to allow xvid to surpass divx in quality. in other scenes the opposite is true. you can check out some screenshots taken in virtualdub (=> no postprocessing) here from my elaborate, web-design-award winning homepage (http://members.fortunecity.de/kilg0r3/index.html) (ca. 1mb, rar 3.0 - sorry for the pop-ups).

again, this is not meant to be a general statement for or against any one of these codecs. I rather hope you tell me what i possibly did wrong in my xvid encode. for i believe i have quiet probabely screwed the encode by using some stupid altcc setting, and, i would really love xvid to look better.

UPDATE

hi! i've updated the 'site'. there are now two archives which also contain the screenshots from the sourceframes.

cheers kilg0r3

--------------------
still just a beginner

[Teegedeck]

...this reminds me that we STILL haven't reached any consensus on a recommendable curve-type and distances!

For goodness' sake, some people here have decently fast PCs - can't you do us a favour and take on the tedious job of encoding the _same_ movie at different settings and pass us your qualified subjective views? I'd really like to do it myself but there's just not so much encoding I can do in my limited computer-time. Else I'd vote for hiding the two-pass-settings deeper in the configs - defaults give something quite near to fluctuating around a constant quantizer and that's not bad at all for everyday use.

[LotionBoy]

tell me what settings you want tested and I'll try to get around to running them on the dual-Athlon.

LotionBoy

[Acaila]

I once tried out 8 different Alt-CC settings on the same movie. My conclusions were that NOT using the Alt-CC gives you equal or better quality than any settings I tried. There was hardly any difference in quantizer distribution between the various settings btw.

There are much more rewarding settings you should mess with, like min/max quantizers, resolution, Lumi Masking and the 'normal' Curve Compression. Just to name a few.

But I doubt this is what you wanted to hear....


Ps.

These are the settings I tested:
Type High Low Strength
Medium 300 100 25
Medium 300 300 50
Medium 100 100 25
Medium 200 100 25
Medium 300 100 25
Medium 100 200 50
Medium 200 100 50
Medium 100 200 50

All I'm trying to say is, if you want to you can tweak XviD's Alt-CC settings to your heart's desire, but it won't give you much better quality than tweaking other 'easier' settings.

[Teegedeck]

Hello Acaila,

I'm not surprised to hear it, actually, as you certainly know very well how to handle traditional CC. :) I can't agree that there are no differences on quantizer-distribution, though. Probably you haven't noticed much because default 'strength' keeps the most horrible of effects from happening - good thing, too, and its main advantage over traditional CC. You actually can't ruin much if you leave strength at default value.

But we still don't know whether there is any virtue in having all these different settings we can use. If you set strength higher, effects of different curve types get very obvious to me. My own limited experience ammounts to little more than 'using types other than low agression and 100 low, 300 high gains you little but might cost you much'.

LotionBoy, if you find you have spare time, you could choose one typical action movie and one drama (clean sources) and try the different curve-types for _identical_ distances (we seem to agree that small frames should be favoured a bit, but I think the widespread 50-200-setting is a bit daring, so why not use 100 low, 300 high? ;)) 'Strength' is another variable, I think, but perhaps you should leave it at a fixed value for starters. Maybe '60' would be good to make the effects of curve-types more visible. Thanx!

[EDITED]
"All I'm trying to say is, if you want to you can tweak XviD's Alt-CC settings to your heart's desire, but it won't give you much better quality than tweaking other 'easier' settings."

Agreed. That's why I'd opt for 'hardcoding' the universally 'safest' settings and hiding the options.

[kilg0r3]

any comments about the apparent superiority of Divx over Xvid? Has any body checked the screen shots?

[sffubs]

I agree that it doesn't usually make much difference, but there is the odd movie that really benefits from messing about with alt-cc.

For instance, Flash Gordon has some very abrupt, extremely high-motion scenes towards the end, which by messing about with alt-cc I managed to get to be pretty much block-free.

I think the option to change some of these settings is quite a valuable feature, which gives this codec an advantage over its competitors.

Maybe you could consider just setting the safe settings as defaults, and advising people to leave them alone if they don't have the patience to mess about with them?

I'll see what I can do about running some tests on my twin AthlonMP box as to how much effect these settings have, but it'll have to wait a week or so (until my exams finish :D)

Cheers,

sffubs

[Teegedeck]

Sorry, kilg0r3, that was like hijacking your thread. Actually I can't have a look at screenshots, only got a very old version of MacUnRAR here. Maybe DivX simply is better? :)

No, we can't accept that...!

@sffubs: how did you change your settings for movie?

[Koepi]

I messed around a little with alt-cc and came to the conclusion that "low" curve aggression gave better results subjectively than medium curve agression.

The rest of the values were defaults...

Regards,
Koepi

[unplugged]

Originally posted by kilg0r3

to my eyes and surprise, the divx codec seems to provide more details and less artifacts with these settings. only in dark scenes, lumi seems to allow xvid to surpass divx in quality. in other scenes the opposite is true.

1. DivX5's B-Frames in motion areas often/always look ugly or unacceptable to be usable. So I feel strongly to not use B-Frame support with this codec, IMHO.

2. XviD can suffer a lot by a *bad* or extreme setup (45 and/or 200 as low and hi totally destroys CC dynamics :rolleyes: ).

3. Medium CC with High 350, Low 99 and Strength 50 has given me good well balanced and pretty "detailed" results (archived with 2-CD 2-hour movies as Matrix 720x424 and The Mummy 704x424).
Making a comparison with DivX5 tuned with your same setup (but with light-psy) XviD has "won the battle" in almost all frames: especially in very low motion frames and very high motion frames :D (great fidelity).

4. It's better to use MPEG quantization for quality encodes, instead of H.263, personally I can see difference between these with full-res/sharp movies (720xYYY).

5. KEEP THE FILTER BAR TO THE LOWEST FILTERING EFFECT IN XVID PLAYBACK FILTER PANEL TO *SEE* THE DETAIL FIDELITY OF THIS CODEC.
Raising image filtering to activate deblocking and/or deringing blurs blurs and blurs the image by UNACCEPTABLE amount :rolleyes: (detail destroyed).
Plus, I warmly advice to NOT change image brightness level from video card overlay panel with NVIDIA boards, use XviD control instead, I am almost sure to have discovered that the first (nvidia overlay) seems to cause visible color (chroma) alteration.

[Teegedeck]

Hello Koepi - another very qualified vote here... I'd think low aggression should lend itself especially well to 'problematic' movies, as it is quite immune to 'jumps' (steeply rising quantizers) in quantization and that's why I prefer it - playing it safe. That's why I was asking sffubs about Flash Gordon,too. But still, maybe 'medium' has advantages when it comes to more run-of-the-mill-type of movies?

[EDITED]
"4. It's better to use MPEG quantization for quality encodes, instead of H.263, personally I can see difference between these with full-res/sharp movies (720xYYY). "

I strongly agree on the superiority of MPEG-quantization - IF you reach compressibility-values of 60% or higher in a test AND use low aggression.

[-h]

I haven't any encodes over 3 minutes in length with XviD, so I can't really comment on curve techniques.

It's possible for DivX5 to look better than XviD at equivalent bitrates, but it's subjective enough that many people will find the opposite to be the case. I don't personally like the look of B-frames with their quantizer doubled, but lowering the B-frame quant multiplier through the registry wipes out the file size benefits that B-frames gave anyway. I haven't found GMC/Qpel to be effective enough to make much difference in their current implementations.

But in terms of which codec I'd encode my own (theoretical) movies with, I'd go with XviD.

-h

[Teegedeck]

Now, that's reassuring... ;)

[kilg0r3]

(1) i agree that mpeg quantizers are the favorable choice when encoding with high bitrates but i don't know if the same is true for lower bitrates like 850 kbit/s asb or in my encode above. probabely modulated would be a good choice then.

(1b) at the above specified bitrate it seems to me that xvid has less detail AND more artifacts than divx eventhough i used h.263 quant.

(2) did anyone have look at the shots?

(3) the shots are not (or at least not exclusively) b, p or keyframes. actually. i don't know what they are. i took them loading a avisynthscript with the line 'select range (12000,1)' into virtualdub and then saved them as image sequence. hence, there also was no filtering involved. so they are all randomly selected.

(3) regarding the future test encodes i#d say we should have at least one example encoded at very high bitrates of about 1400-1900. For,IMHO, xvid can definitely make better use of high bitrates than divx5, which still tends to blur. however, i am not sure about that for i, stupidly, used b-frames in my last highbitrate encode with divx.

(4) is there a way to tell which frames are b and which are p-frames

i'll substitute the files on my 'site' with zip-archives for better compatibility.

[Nic]

I recently did True Romance with XviD & DivX5...DivX5 won hands down
:(

Dont know why, on the film Ronin, XviD beat DivX very easily (especially on the car chases (good Motion code :) ))...So I think it does vary.

Hmmmm...Its about time I did a release with Syskin's motion code turned on isn't it.....hmmmm :)

-Nic

[Teegedeck]

If we're talking 1-CD encodings, maybe it's all due to B-frames' space-savings?

*Sigh* I hope gruel finds time to work on that new motion-search... Summer is indeed a bad season for open-source-development...

[kilg0r3]

yes, i think it would be great to see some dual-athlon people and some developers putting there heads toghether in order to create a series of systematic test encodes. if necessary, the design of such a test could previously be discussed in a new thread here in the forum.
a tool which is able to print the codec settings to a file would be very handy in this context.

.... now shutting up ....| :rolleyes:

[Acaila]

yes, i think it would be great to see some dual-athlon people and some developers putting there heads toghether in order to create a series of systematic test encodes.
Just buzz when you need me :)

[LotionBoy]

Originally posted by Acaila

Just buzz when you need me :)

yeah, me too.

I'll try and throw something together this weekend. I'll be in and out of my house, so maybe I can just queue some stuff up while I'm out enjoying the beach and the sun.
SO you want me to try tests at different agressions with the same values for low and high. Can do. 1CD rip.
CTHD is my fav. for action movies, though it's a little long.
Anyone know what sort of print sex,lies, and videotape is? just grabbed a copy of that and I could use that for the drama. I'll see what I can do. Feel free to post more suggestions of what I should encode.

LotionBoy

[MoonWalker]

a tool which is able to print the codec settings to a file would be very handy in this context.

I don't know how usefull is but I have made a tool tha analyzes the Debugoutput of the XviD codec...

[MoonWalker]

Hi,

I just got a copy of The Matrix and I am willing to run some test's (encoding the whole movie) for XviD's Alt-CC...Could you tell me the setings you want to use to alt-cc..And how many cd's??

MoonWalker

[Emp3r0r]

i just redid my matrix to test divx 5 and i tried all the new pro features... I saw weird interlacing type problems on the encode (most likely forced film got unchecked). I haven't had a chance to go back and double check the settings I used but overall the quality wasn't too bad looking... I think I'll stick with XVID anyway cause I'm more familiar with it and it produces acceptable quality everytime. I was very pleased when I fit Blow (2+ hrs) on 1 CD with 640 resolution and it still looked good (Mode2 helped). I just started Matrix again, i'll post my results.

[Teegedeck]

Originally posted by MoonWalker
Hi,

I just got a copy of The Matrix and I am willing to run some test's (encoding the whole movie) for XviD's Alt-CC...Could you tell me the setings you want to use to alt-cc..And how many cd's??

MoonWalker

Theoretically reasonable distance-settings would be the defaults or high 300, low 100 if we agree to put special emphasis on small frames - but in no case an extreme setting like low 100, high 100 - definitely only 'strength' saves encodes for such settings.

One CD is of course more of a test than 2 CDs (well, it's just about aiming for a bits/pixel-ratio that leads to the use of quantizers higher than '3'. :)

Then, how about setting strength to '60' and testing the different agression-types? And perhaps you'd better not use lumi-masking on the Matrix. Thanks for volunteering...:D

[MoonWalker]

OK..I will aim for 1CD sound at 128Kbps and the different aggresion types..

I will post the results soon..:)

BTW nice avatar Teegedeck :)

MoonWalker

[Teegedeck]

Originally posted by MoonWalker

I will post the results soon..:)

BTW nice avatar Teegedeck :)


Glad you like it... Originally it even is an animated GIF...

[Emp3r0r]

wow, my matrix rip finished and i seemed to have topped out on quality with Divx5... loooks great and much crisper than Divx4... yet not too far ahead of Xvid.

[Acaila]

Since we're all hooked on The Matrix, I think I'll run some CC-tests on it (someone else can do the Alt-CC tests :)).

I'm currently queing up 36 2nd passes, I'll report back when they're done.

[kilg0r3]

jsut wanted to mention this: make sure you compare clips of film and not just screenshots as i did in my first post. for when i actually wathed the movies the divx5 clip had severe problems with moskito noise and the moving-background problem. both, you couldn't see on the screenshots.

->

[MoonWalker]

I have finished the test with Matrix..You can see the results here..http://forum.doom9.org/showthread.php?s=&threadid=27190

MoonWalker

[unixfs]

Hi, I want to share my results and ask you expert coders some suggestion regarding xvid.

I wanted to compare divx502 and xvid (XviD-22052002-1 by Koepi)
at low resulution
so I encoded Titanic(190 mins) and Pearl Harbour (175 mins)
in order to fit them on a single CD each.

The setting were the following for both:
- Divx502: two passes, B frames only, 448 kb/s, no pre- or post-processing
- Xvid: two passes, Quant. matrix: modulated, lumi-masking enabled during the second pass,
payback with automatic bias when used with cc, default setting when used with the ordinary cc (I made two second passes to compare them); desired file-size: 620000 KB

Program used: Dvd2avi 1.86 by Ogo

Letting alone audio, both movies encoded with divx looked good when watched from half a meter, and very good when watched from farther away (as when I watch the TV); besides both files where as large as expected.

With xvid, instead, both movies where 800 MB, and when played (I tried both divx decoder and ffdshow, with all the possible settings of post-processing) they appeared very very blocky, really unwatchable.

Since I read so many good things about xvid, I'm sure I set the wrong parameters, but don't know which. I tried to attain as much as possible to Koepi's guide.

Your suggestions will be very welcome.
Thanks.

[Teegedeck]

Ho-humm. Try lumi-masking in _both_ passes, alt.CC with low aggression, low distance 100 and high distance 300, strength 50%, I- and P-frames 2-12. This ain't a magic trick, just my standard setup that has never disappointed me.

[unixfs]

I followed your suggestions and encoded a shorter movie with the settings you suggested plus I-frames Qs: 1-6.

Now it's very much better, but there are often pink areas around the perimeters of figures, and small objects such as hands in the background are very fuzzy. The file is undersized: 450MB instead of 610.

Misregard these problems, now the movie is much closer to it's equivalent made with divx5.

Is the bitrate setting in the alt-cc panel an average bitrate parameter?

Thanks again.

[Teegedeck]

Originally posted by unixfs


Is the bitrate setting in the alt-cc panel an average bitrate parameter?

?!There's no such thing?!

Perhaps you're talking about the maximum-bitrate? (set that one to '999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999'. :))

BTW, instead of asking right away, please just hold the mouse-pointer over a feature that eludes you and read the tool-tip help...