Xvid made a believer out of me... [Archive]

SleepEXE

20th January 2002, 03:36

Well, Xvid has made a believer out of me. I finally got around to doing some direct comparisons this weekend. I pitted the 01/07/02 build of Xvid against DivX 3.11, both using Gordian Knot to prepare the stats files for 2-pass.

I'm not usually one to make such strong statements, but Xvid simply kicks 3.11's ass. In bright, highly-detailed areas, it's often difficult to make out much of a difference between the two, although Xvid does appear to preserve detail ever so slightly better. Where Xvid really excels, however, is at preserving gradients and detail under low-light conditions...which as we all know has forever been the bane of DivX. I suspected that luma correction in Gordian Knot may have unfairly exacerbated this effect in the DivX 3.11 file, so I re-encoded it with the feature disabled.

Here are some screen grabs for comparision. I recommend downloading them all to your hard drive and quickly toggling between them in an image editor such as IrfanView (http://www.irfanview.com) (great, free image viewer by the way) or ACDSee.

Original frameserved file (http://orca.st.usm.edu/~jmneal/files/gf3_orig.jpg)
Xvid build 01/07/02; Gknot'd (http://orca.st.usm.edu/~jmneal/files/gf3_xvid.jpg)
DivX 3.11; Gknot'd (http://orca.st.usm.edu/~jmneal/files/gf3_div3.jpg)

Marked differences can be noted in at least three places:

1. the walls - the original frame clearly shows a delineation of bricks that compose the wall. Xvid reproduces the bricks faithfully, DivX 3.11 washes them out in a blur.

2. the cabinets - the DivX 3.11 has obvious macroblocking, Xvid displays a nice, clean gradient.

3. the sweater - look at Pacino's shoulder, DivX 3.11 replaces the texture of the sweater with a blotched patch of red, Xvid preserves the texture much more closely to the original

These screengrabs were done in Virtualdub, thus no post-processing is applied. Final file sizes were within 1.5MB of each other (each ~1.1GB, average bitrate 900kbps). This is only a single frame of course, but everywhere I looked, Xvid was at least as good as the DivX 3.11 and usually better.

I know Doom9 is working on a new codec roundup now and I'm interested to see if his results compare favorably with mine. I guess I need to try DivX 4.12 now...

Best regards,
SleepEXE

-h

20th January 2002, 04:18

I honestly didn't think the difference would be that large. Quite impressive.

If you're feeling daring, you could see what the max quantiser from the .stats file was, and enable lumi masking with xvid's max quantiser set to (whatever-the-max-was)+1.

-h

SleepEXE

20th January 2002, 06:34

I'd be glad to do so, but I'm not sure how to find the maximum quantiser unless it involves custom scanning the stats file...

Incidentally, I posted a message (http://arstechnica.infopop.net/OpenTopic/page?a=tpc&s=50009562&f=67909965&m=8760912453) in Arstechnica's A/V forum referencing this thread and came under some fire about the quality of the DivX 3.11 sample. I think it's explained adequately there, but it's worth mentioning that my comparison isn't intended to represent the best possible with either codec, simply that Xvid seems to do a better job given similar constraints (i.e. 1-2 CD rips).

Best regards,
SleepEXE

gldblade

20th January 2002, 17:15

Hmm... The XviD one looks a little grainy. Looking at the bottom part of the wal...

cofferscuffs

20th January 2002, 17:22

Open up both piccies, focus on the hair and flick serveral times through both of them... XviD is MUCH sharper. The grainyness I can live with.... is it time for me to jump ship to XviD??

gldblade

20th January 2002, 17:30

I think the graininess of the image is ok also, but if it changes during playback (such as noise), I will probably notice it.

Tri

20th January 2002, 18:03

em, where can I get a compiled version of Xvid?

Or can someone tell me how to compile it with VS 6?

SleepEXE

20th January 2002, 18:23

I think the graininess of the image is ok also, but if it changes during playback (such as noise), I will probably notice it.

The graininess is present in the source material, so the fact that Xvid reproduces it is a good thing in my opinion. Playing through that section of the movie, the DivX 3.11 version exhibits more of the macroblock "crawlies" on the wall than the Xvid...I think that's what you're describing. About jumping ship to Xvid, for me that's going to be a hard decision to make. There's the comfort, reliability, and known performance of DivX 3.11, but I'm leaning very hard toward Xvid now. The Xvid encoding of "The Godfather 3" is definitely a keeper. I don't want to jump to conclusions until I've tested it more extensively with material such as anime (a small portion of what I do, but important nonetheless).

Tri, you can grab the latest builds at Umaniac's site (http://www.heimsnet.is/kg/).

Best regards,
SleepEXE

MaTTeR

20th January 2002, 22:09

@SleepEXE

Thanks so much for the insight. I'm now drooling and getting very excited about this new codec:D

-h

20th January 2002, 23:15

I'm actually interested that it retained that detail - I've seen such noise removed by quantiser thresholding in the past. Which quantiser were you using?

I wonder what kind of results doom9's encoding will turn up.

-h

Doom9

20th January 2002, 23:38

aargh.. nandub just crashed before the end of my spr divx3 first pass... another few hours wasted.. it will take some time getting all the files right.. right now I have everything for the 1 cd test and spr in divx4

SleepEXE

21st January 2002, 01:49

-h, I simply left the quantiser window at the default of 2-12. Other settings were keyframe 300, Motion Search Precision 5, Quantization MPEG, No Luminance Masking. Then prepared the stats file in Gordian Knot following the recommendations to disable Motion Correction and Luma Correction. Altogether, a run-of-the-mill Xvid 2-pass encoding.

Best regards,
SleepEXE

manono

21st January 2002, 09:43

Regarding the Arstechnica exchange-the Emperor (morello) has no clothes. If your ripping methods aren't the same as his, he doesn't want to hear about it, and his are frozen in a time about six months old.

I was surprised to read the suggestions over there to the guy that wanted to rip The Fast And The Furious (before the diatribes about ripping stuff you don't own). This from guys that are supposed to be up to date with the latest methods. Goes to show that if you don't follow Doom9's site, you'll quickly be left behind.

ChristianHJW

21st January 2002, 17:28

Hmmmm .... finally XviD will be tested against DivX3 ( and 4 ) from a real authority in encoding :cool: .
I admit i always was too lazy to do it .... :D ..... looking forward to see the results ...

NeVeRLiFt

22nd January 2002, 10:11

Hmmmmm you got me interested.
Will check into this and see for myself.

NeVeRLiFt

22nd January 2002, 10:58

-h

22nd January 2002, 11:40

Well I did some tests video capturing with this codec ;) its good people really should learn that you can capture straight to DivX and have good quality. Will do some tests encoding and post back.
You know the captures look good but sorta grainy..... maybe thats just how this codec is?

The MPEG quantiser will produce 'grainier' results than H.263, in addition to being much slower. If you're using H.263 at the moment.. then you can try using a higher quantiser :)

-h

Teegedeck

22nd January 2002, 11:41

Originally posted by NeVeRLiFt

You know the captures look good but sorta grainy..... maybe thats just how this codec is?
I'd like to know if the graininess was in the source, too (I'd guess so, after all only a few DVDs are perfectly 'clean')?

If you don't like that effect, you could try using the H.263 quantizer; it's the same that DivX4 uses and could smoother out some details/noise.

Teegedeck

22nd January 2002, 12:17

BTW, some time back there have been plans that XviD should automatically switch between MPEG (for low quantizers) and H.263 (for higher quantizers). Just today someone said he hacked such a feature for personal use and that it works fine... I'm waiting for this, as it could make the codec look much better at low bitrates, I think.

morello12

22nd January 2002, 19:15

Incidentally, I posted a message in Arstechnica's A/V forum referencing this thread and came under some fire about the quality of the DivX 3.11 sample. I think it's explained adequately there, but it's worth mentioning that my comparison isn't intended to represent the best possible with either codec, simply that Xvid seems to do a better job given similar constraints (i.e. 1-2 CD rips).

You came under fire, only because you refused to post the most important fact in the thread. The bitrate. Sorry I didn't spend enough time reading through messages for it, I and most people there rarely ever use bitrates that would show results that your test has. I explained myself, and took back my "attack" once I found out you were testing lower bitrates, and your conclusion does not apply to something I would encode.

Regarding the Arstechnica exchange-the Emperor (morello) has no clothes. If your ripping methods aren't the same as his, he doesn't want to hear about it, and his are frozen in a time about six months old.

First of all, it's really low to attack me from another forum that I don't regularly frequent. Secondally it's even more disgraceful to make me out as something I'm not, some sort of "emporer" if you will.

I will admit, I don't have time to refine my methods every week. I tend to find something that works really well and stick with it. I really don't rip movies anymore, so I don't give as much input on the area as I used to when Mpeg4v2 was promenient.

Let's lay off the attacks ok? If he posted somewhere in his write-up on ArsT that your reaction might not be "WOW!" if you already use higher bitrates then I wouldn't have tried to shoot it down so quick.

I was surprised to read the suggestions over there to the guy that wanted to rip The Fast And The Furious (before the diatribes about ripping stuff you don't own). This from guys that are supposed to be up to date with the latest methods. Goes to show that if you don't follow Doom9's site, you'll quickly be left behind.

ArsT's A/V forum is more focused on A/V hardware and problems / TV capping than it is about ripping and encoding. It's not a specialized forum on ripping/encoding like this, so don't try to make it out as one, as you come off ignorant.

I frequently refer people here, as I know this place is much more dedicated to that material than ArsT.

SleepEXE

23rd January 2002, 00:41

You came under fire, only because you refused to post the most important fact in the thread. The bitrate. Sorry I didn't spend enough time reading through messages for it...

I did most certainly state the "most important fact" in my initial response, and I quote:

Final file sizes were within 1.5MB of each other (each ~1.1GB, average bitrate 900kbps).

I and most people there rarely ever use bitrates that would show results that your test has.

...which obviously invalidates any merit of my test. I plainly indicated on more than one occasion that it was a comparitive test between DivX 3.11 and Xvid, and gave very specific conditions under which the tests were conducted. Frankly, I think your response was more appropriate to someone who provides no substance and simply says "Wooohooo, Xvid is the baddest mutha' on the block. I pity the fool that's still using DivX 3.11".

I explained myself, and took back my "attack" once I found out you were testing lower bitrates...

For what it's worth, I was never angry at you nor did I take your initial response as an attack. A bit abrasive, perhaps, but that doesn't bother me. I harbor no ill will.

...and your conclusion does not apply to something I would encode.

Perhaps not, but given all the cogent information, shouldn't people be encouraged to draw their own conclusions and debate the merits of each codec? And even if it does not apply to something you would encode, doesn't it stand to reason that if Xvid can outperform DivX 3.11 at low bitrates, it probably will do so at higher bitrates? After all, in the spirit of researching new codecs, we want the best possible quality for the file size.

Best regards,
SleepEXE

NeVeRLiFt

23rd January 2002, 00:45

Originally posted by Teegedeck

I'd like to know if the graininess was in the source, too (I'd guess so, after all only a few DVDs are perfectly 'clean')?

If you don't like that effect, you could try using the H.263 quantizer; it's the same that DivX4 uses and could smoother out some details/noise.

You know the source was kinda grainy.

I would like to say this XVID Codec is very good!
Its in Alpha stage I belive and its already this good and bug free!
I mean I have had no stability problems with it yet and that cant be said for DivX4.xx

One other thing, have you guys heard about MS teaming up with HomeDVD makers and making a HomeDVD unit that supports mpeg4/WMV ? Its supposed to come out this year or next.
Will we be switching codecs?

Teegedeck

23rd January 2002, 01:00

Originally posted by NeVeRLiFt

One other thing, have you guys heard about MS teaming up with HomeDVD makers and making a HomeDVD unit that supports mpeg4/WMV ? Its supposed to come out this year or next.
There will be players by Panasonic, I've read.

Will we be switching codecs?
For WMV? :eek: You've gotta be joking!

morello12

23rd January 2002, 01:07

To clarify one last thing SleepEXE, I do not want to argue with you, and I do believe finding better codecs is for the greater good.

When I said what I said that invoked this response:

I did most certainly state the "most important fact" in my initial response, and I quote:

quote:
--------------------------------------------------------------------------------
Final file sizes were within 1.5MB of each other (each ~1.1GB, average bitrate 900kbps).
--------------------------------------------------------------------------------

I was referring solely to the thread on Ars. That was what I read. Then I clicked your link and looked at the pictures you had. I am sorry I didn't read your entire post in Doom9's forums, or I would have noticed what you said about the bitrate at the bottom. It was a misunderstanding, as you could have posted the relevant info on the thread in ars, not linked to it, and I could have read your other post here more carefully.

Ah well.

On another note, I think soon the most sought after feature in any video codec will be it's ability to embed the codec inside the final file, because there's becoming way too many different ones.

Doom9

23rd January 2002, 01:34

switching to wmv? over my dead body.. wmv8 looks worse that both divx4 and sbc (as you can see in my old comparision).. and encoding takes ages.. my computer (athlon xp 1800) was working from 7am till 1am to do a 2 pass encoding of saving private ryan.. I could do 3 divx4 rips in the same time.. and quality would be better

NeVeRLiFt

23rd January 2002, 03:33

So this is not a good thing. These new HomeDVD units will not use anything but WMV? Is there any hope of them supporting other mpeg4 codecs?
And Doom9 your more than right about the quality of WMV. Not only that there are no tools or software to use with it.
When I read that news about this.... I had a dream of my DivX rips playing in my HomeDVD unit! Damn! Damn!
/me goes cries

Doom9

23rd January 2002, 11:32

so far only 1 or 2 players with wmv support have been announced and nothing can be bought yet. divxnetworks is trying to convince manufacturers to support their codec (though I think they'll eventually go for iso mpeg4 compatibility as there the supports seems to be much larger). I don't think you'll ever see divx3 support in professionally done units because of the copyright issue... iso mpeg4 compatible seems to be possible though (but requires that we finally dump divx3.. and at the current point there's no fully iso compliant mpeg4 codec that really beats divx3 hands down in every case which makes that step rather painful)

Teegedeck

23rd January 2002, 11:42

Originally posted by Doom9
and at the current point there's no fully iso compliant mpeg4 codec that really beats divx3 hands down in every case which makes that step rather painful)
[looking for hints on the upcoming codec-comparison] So DivX 3.11 still beats all others on 'The Matrix'?

NeVeRLiFt

23rd January 2002, 21:56

Originally posted by Teegedeck

[looking for hints on the upcoming codec-comparison] So DivX 3.11 still beats all others on 'The Matrix'?

;)

NeVeRLiFt

24th January 2002, 03:41

http://www.eetimes.com/story/OEG20011212S0060

Had a guy give me this link.

morello12

24th January 2002, 06:14

The specs to MS's new codec certaintly look impressive, and if it can deliver 1920x1080 @ 60 fps like it promises, I can see a lot of people with money converting to it.

There'll be about 10 anti-MS people for every 1 looking at things logically and unbiased (in online discussions), but if MS plays their cards right this could be something good

-h

24th January 2002, 07:14

Sadly, I doubt we'll see anything revolutionary from MS. WMV7 and WMV8 were huge disappointments - the only difference was the image prefilter and a big slowdown (despite whatever was claimed on their propoganda site).

-h

morello12

24th January 2002, 16:04

I can't say I agree with you. This is not just some update bragging better compression, this is the entire shabang.

Afterall, MS was the one who released Mpeg4v2 and that was the best thing since sliced bread, it was ages ahead of it's time. I think they might pull a nice surprise with this new codec.

And the audio specs are quite nice too, 24-bit 96khz (it might have even been higher).

Teegedeck

24th January 2002, 17:04

I think there's no sense in speculating about something none of us can possibly have tested yet - let's wait and see.

One thing that REALLY, really annoys me about commercial codecs, is the marketing included. Everything new from M$ get's announced by Uncle Bill as, right, 'the best thing since sliced bread', people from DXN bragging on the whole day long about how fantastic and revolutionary DivX.4.5 and 5 will be - I just can't stand that anymore! Let them deliver something first and THEN we can talk about it with a little more meaning.

Killah

24th January 2002, 18:00

I agree with Teegedeck... I think there is a lot of politics involved in this mediaformat thing... Ever since Quicktime came out, M$'s marketing machine is working overtime making their equivalant a succes (VFW then DirectX)...

Doom9

24th January 2002, 19:10

well.. in DXn's defense.. they haven't really been hyping it. betaboy mentioned something in these forums a while back but that's been it... until now we're all waiting to see something new (that includes me).

I've already disproved the MS hype about their new codecs twice (wmv7 and wmv8) and without divulging any results from the upcoming comparision.. just check my latest comparision to see how well wmv8 performs. It was the leader of the pack for automatic 2 pass movie creation when it came out, but divx3 already had mm4 which when properly done could probably have lived up to wmv8, but since nandub wmv8 no longer reigns and shouldn't be used by anyone except for doing a codec comparision ;)

-h

25th January 2002, 00:26

I just think that, given the *insignificant* differences between MP42/MP43, WMV7 and WMV8, I'm not so sure that any codec revisions will occur in Corona. The biggest difference will be the infrastructure surrounding the video delivery.

The audio improvements, while welcome, don't fill me with confidence either. "WMA8 is CD quality at 48 kbps.."

I'm just judging MS by their past performances in video/audio. The MP4x codecs were revolutionary - granted - but they seem to have fired the engineers responsible.

-h

Stabmaster-Arson

25th January 2002, 00:27

What I wanna know is, Is Divx 311 REALLY the king, or is it just becauase of Nandubs utilisation of it. Effectively, because of nandub, 311 is obviously the best so far hands down, but what if there was a modified version of nandub, set to use other codecs.. then it would be fair testing ground.

Stabmaster-Arson

25th January 2002, 00:31

Dont forget too, without some really funky players, our DVDivX rips are gonna be no good for a home unit, even if they suported divx311 which they wont. they been inverse telecined and are 24 fps now......

Doom9

25th January 2002, 00:31

well... other 2 pass codecs do pretty much what nandub does for divx4.. no need for a modified nandub. nando was working on integrating opendivx support, but then divx4 came and it added 2 pass support, and nando stopped development (at least as far as I know).. in any case, the point is nandub isn't so much different from other 2 pass procedures

-h

25th January 2002, 00:48

Just wondering - would you prefer the GKnot functionality to be merged into XviD itself, so you just set a bitrate for the 2nd pass and it goes about its merry way?

This might make editing the stats with Gknot a bit more complex though. Is there much in Gknot you use with regards to XviD?

-h

Stabmaster-Arson

25th January 2002, 02:20

i have to post here about this. I ran into a MAJOR bug with Xvid. When using 1 pass CBR, at almsot ANY bitrate, it would totally ignore what i chose and encode super low, like 45K/sec. 1 minute was 2.5 MB and looked like crap. This is windows 2000 Sp2. even at 6000 bitrate it was 2500K for 1 minute, and super lousy quality as should be expected. same esult at 1511 kbit too.

Stabmaster-Arson

25th January 2002, 02:22

oh and i was using vdub 1.4.7 .

Stabmaster-Arson

25th January 2002, 02:27

Im gonna try some more, the only different option i chose was on the force optimisations page, by defailt it selected MMX, I click SSE as well (p3 800) im gonna uncheck sse and see what happens.

-h

25th January 2002, 02:34

I thought the CBR undersize issue was fixed a while ago.. guess it's reappeared? I'll have a play.

-h

Teegedeck

25th January 2002, 03:22

Yes and no. Personally I'm happy with it as it is. But as we - to make things easier for former DivX4-users - will get an integrated 2-pass, anyway, I would really like to see Gordian Knot's code included into XivD's 2-pass, as I believe its method of scaling all frames alike but with special attention to a minimum bitrate works best of all methods.

But GKnot is more - as I think Doom9 called it a 'swiss army knife'. Most importantly, you can open dvd2avi-projects and influence the bitrate of specific sequences - I don't know of any other software that can do that. So while for most users an integrated 2-pass-funtionality in XviD by no doubt would be a big relief, XviD still should produce .stats-files that can be manipulated. The problem at the moment would be that inside of Gordian Knot you can't push the bitrate of certain scenes without calculating the new stats, so this would get in the way of any calculation inside of the codec. Perhaps we should ask TheWEF whether he can do something for us? (I hardly dare because he must be one of the busiest men in the business ;))