Oh, NO!!! Not another Divx5.1.1 Quality vs. Xvid Quality Thread!

[ungeraynrand]

Of course, 'Quality' is subjective but certain features that 'Quality' are judged upon are not. Mostly, I am interested in your observations on whether or not Divx5.1.1 produces sharper images than the previous version (5.0) versus Xvid. What's your take on this matter? Whatever enchancements there are, they are theoretical and not noticed by me at least.


So? Divx sharper or Xvid sharper?

in the previous versions Divx was less sharp and slightly blurry, some ppl prefer this, that's fine. i prefer that divx preserve as much detail as possible.

[FreakNGoat]

I am currently encoding Lawnmower man with XviD VHQ 2 and Qpel on. I've already encoded the same movie using Divx 5.1.1. In the Divx version, I've noticed that using Qpel produces drastically better results than using the highly-touted "Slowest" mode that people are raving about. At least in the case of this movie, Slowest mode was literally a big waste of time (took twice as long to encode compared to Qpel).

I've been using Qpel since Divx 5.02 and most of my rips seem very sharp; it mostly depends on the resolution I encode them at. It is almost immediately obvious to me when Qpel has not been used; the picture has that smooth/blurrly quality to it. I haven't done much with XviD however so I'll post my results after this movie is done encoding.

[DigitAl56K]

QPel has a number of drawbacks, including higher minimum decoding requirements and incompatibility with DivX Certified hardware devices.

[ZZZERO]

Originally posted by FreakNGoat
I've noticed that using Qpel produces drastically better results than using the highly-touted "Slowest" mode that people are raving about. At least in the case of this movie, Slowest mode was literally a big waste of time (took twice as long to encode compared to Qpel).


Not knowing anything about Qpel I have to ask why is this an either/or situation? Wouldn't encoding with Qpel in slowest mode be best of all then?

[bond]

I always use QPel in XviD as it produces a much sharper image without hurting the compressibility that much (and i dont care about "DivX Certified" hardware devices :p )

in fact it seems to me that only because of the lack of hardware support for QPel at the moment DivX5 tries everything to get people away from qpel (make it only work with Standard mode, hiding the feature in their interface, i am not sure if they even work on their qpel implementation to make it better?...)

i can only say: people who dont think about buying a standalone in the next 1 or 2 years should use QPel, it is really worth the quality increase!!!
maybe it doesnt even take that long to get standalone support for qpel when the first people will complain that they cant playback all of their divx movies in "divx certified" hardware devices...

[DigitAl56K]

QPel is only useful at higher bitrates, at low bitrates it can actually impact negatively on quality.

[bond]

Originally posted by DigitAl56K
QPel is only useful at higher bitrates, at low bitrates it can actually impact negatively on quality.hm, rarely, i only make 1cd rips...

on high bitrates there are already a lot of details, qpel brings details on medium bitrate ranges (1cd)

[cordraconis]

sorry to butt-in on this, but when I recently encoded Gladiator, I first did a Qpel-encode, and some low-motion scenes looked incredibly crisp. (Like the scene where Marcus Aurelius looks over the battle while seated on his horse.)

But mostly the Qpel introduced artifacts (tiny, but still), like the feet of the horses while they charge.
Maybe this was because of the desaturated look of the first 20 minutes, or maybe because of the high bitrate (2CD's), but eventually I decided to forget the Qpel, use very mild C3D settings, and a lot of multipasses.

Yes, I lost 4 or 5 "supercrisp" scenes, but personally, the rest of the movie looked better.

Maybe it's better in the 5.1.1 version, but I haven't tried it (yet).
So far my experiments/experiences on Qpel...

[Enigmax]

Ground to use both codecs, DivX and XviD. But I must recognize that XviD surpasses in quality to DivX.
An image much more defined and with less macroblocks.
XviD is best quality, but DivX is easy.

EDIT:In dev-api-4 of XviD, the obtained quality using QPEL and Quants MPEG is insurmountable.

Greetings

[ZZZERO]

Originally posted by bond
in fact it seems to me that only because of the lack of hardware support for QPel at the moment DivX5 tries everything to get people away from qpel (make it only work with Standard mode, hiding the feature in their interface, i am not sure if they even work on their qpel implementation to make it better?...)



Is the part about QPel only working in standard mode still true for 5.1? I see nothing about that in the divx codec guide on the divx website.

[FreakNGoat]

Qpel increases quality even at low bitrates for movies that compress decently. I always use it when I'm encoding a movie above 512 pixel horizontal resolution and ~60% compressibility check in GordianKnot (for Divx 5.0x... I haven't used 5.1.1 long enough yet to find the ideal compressibility point). The majority of my rips are 1-CD. At the conditions I stated above, the artifacts introduced by Qpel are minor and are not noticable to me. In fact, the artifacts introduced without Qpel are in fact far greater--blurriness and loss of detail.

Yes, for movies with poor compressibility that you want to fit on a single CD, Qpel introduces too many artifacts and at low resolutions crispness cannot be attained anyway.

As far as hardware support, I'm going to go out on a limb here and say I think most people on this board do not care about this. Those who don't care should use Qpel, those who do should not. Many people on this board are using OGM, Matroska, XViD, etc. Formats which may have no chance of hardware support. It does seem that Divx is trying to get people away from using Qpel, which is unfortunate. Not creating a profile for those who don't care about hardware support seems odd. Most people I'd say also have fast enough system to decode Qpel.

When we're talking about highest possible quality encoding, it does not make sense to introduce the limitations of hardware support and slow system speeds which only effects the minority of people.

Is the part about QPel only working in standard mode still true for 5.1?

Yes, and 5.1.1 as well.

EDIT:In dev-api-4 of XviD, the obtained quality using QPEL and Quants MPEG is insurmountable.

I'll have to check this out. Comparing Divx 5.1.1 to dev-api-3 shows Divx the winner. Ha, I realized I was using dev-api-3 when I was encoding a movie last night for comparison; I'll have to get dev-api-4.

[ZZZERO]

For those with lots of experience using Qpel, how does having GMC on or off affect the results?

[bond]

Originally posted by ZZZERO
For those with lots of experience using Qpel, how does having GMC on or off affect the results?i dont use it as, once i tested it (dev-api-3) it didnt gave me a nice result (didnt test it in dev-api4), but doom9 for example uses it
of course you will have to differentiate also between xvid and divx5 as qpel/gmc implementation is not qpel/gmc implementation (as i wrote before i wonder if divxnetworks still works on their qpel/gmc implementation...)

Originally posted by FreakNGoat
When we're talking about highest possible quality encoding, it does not make sense to introduce the limitations of hardware support and slow system speeds which only effects the minority of people.yup, thats my opinion too

in fact i am waiting for the day when a hardware manufacturer will listen to all the angry customers, who are not able to playback all their divx movies in these funky "first generation" hardware devices, and will release a real mpeg-4 compatible standalone player :)
(perhaps this will take 2 years, perhaps not, but i am sure it will come when people will ask for it, and i am sure they do)

[mikeson]

@FreakNGoat:
Many people on this board are using OGM, Matroska, XViD, etc. Formats which may have no chance of hardware support
I think that ChristianHJW will not agree with you on this (neither will I). :p Anyway where did you get XviD don't have hardware support? It's fully MPEG4 compliant... :o
Comparing Divx 5.1.1 to dev-api-3 shows Divx the winner
Opinions like this are irrelevant as you haven't prove it. You shouldn't say things like this on this board.

About crispness in XviD and DivX. AFAIK DivX is still using H.263 quantization type, in XviD you can choose between several of them (and MPEG quantization type will always be more sharper than H.263). So in the end with XviD you can attain more sharper image IMHO.

[FreakNGoat]

I think that ChristianHJW will not agree with you on this (as I won't). Anyway where did you get XviD don't have hardware support? It's fully MPEG4 compliant...

I'm not saying XviD won't have hardware support, or OGM or Matroska for that matter. Divx Qpel, which was the issue here, is also MPEG4 compliant, is it not? They just aren't "guaranteed" like the Divx certification scheme. By using them, you probably have higher priorities than maximum compatibility. Anyway, this is a generalization, and obviously not true in all cases.

Opinions like this are irrelevant as you haven't prove it. You shouldn't say things like this on this board.

Yes, you are right. I did not mean to make a broad sweeping statement like that. After reading it I realize I did. In my opinion, comparing Divx 5.1.1 to Xvid dev-api-3 in the movie I encoded the other night showed Divx to be higher quality. This is of course one sample; I realize to truely show one format better than another many samples are required. Later today, I'll try to post some clips.

[bond]

in fact this "Divx certification program" is something like an own private standard created by divx-networks, focusing, what else, on divx5 (and divx3)

there is a good reason why industry-wide standards like mpeg-4 (not created by one single company) exist:
for example if a hardware manufacturer would decide to produce a mpeg-4 compatible standalone player, you can be sure to be able to play back every mpeg-4 compatible codec, including xvid!

[FreakNGoat]

Originally posted by mikeson
About crispness in XviD and DivX. AFAIK DivX is still using H.263 quantization type, in XviD you can choose between several of them (and MPEG quantization type will always be more sharper than H.263). So in the end with XviD you can attain more sharper image IMHO.

There is a major flaw in that argument. MPEG quant may be sharper than H.263 quant, all other things being equal. However, in comparing codecs, this is not the case. Other things contribute to sharpness as well, so in the end sharpness depends on more than simply which quant type you use.

[mikeson]

@FreakNGoat:
I'm talking about situation when you can benefit from codec posibilities. And I think that you would agree that in this case crispness is about to talk when codec has almost equal features. So when talking about that, what features (so be it, except quantization type) has DivX and XviD doesn't have? GMC? It doesn't matter at all IMHO... And what about VHQ? Chroma motion compensation? Chroma optimizer? Quantizer restrictions? Advanced b-frames settings (at least more advanced than in DivX)? I don't want to start flamewar, I'm just argumenting. I think that in the end, quantization type (and I think that I haven't mentioned custom matrices like hvs-best or andreas, that achieve much more sharper image) comes into play and they play VERY important role on overall image crispness IMHO...

[FreakNGoat]

Originally posted by mikeson
@FreakNGoat:
I'm talking about situation when you can benefit from codec posibilities. And I think that you would agree that in this case crispness is about to talk when codec has almost equal features. So when talking about that, what features (so be it, except quantization type) has DivX and XviD doesn't have? GMC? It doesn't matter at all IMHO... And what about VHQ? Chroma motion compensation? Chroma optimizer? Quantizer restrictions? Advanced b-frames settings (at least more advanced than in DivX)? I don't want to start flamewar, I'm just argumenting. I think that in the end, quantization type (and I think that I haven't mentioned custom matrices like hvs-best or andreas, that achieve much more sharper image) comes into play and they play VERY important role on overall image crispness IMHO...

I also do not want to start a flamewar, but this is a good exchange.

Even if both codecs have the exact same features, this is still not a basis for saying everything is equal, and therefore the codec with the "best" quant type wins. The implementation of these features could be and is most likely vastly different. This is probably what is most important, especially when you are talking about lossy visual compression. The algorithms used have to decide what information to "throw away" and what to keep while keeping the output as close to the original as possible.

Though XviD has many more customizable features than Divx, does this mean that XviD is sharper? Maybe, maybe not. It could be possible that Divx just works better with the features it does have. Because XviD supports supposedly "better" quant types, does this mean it produces sharper output? Perhaps, but saying that it does exclusively because of quant type would be incorrect. Now it seems you have backed off of this argument slightly, to say that quant type plays a very important role on the overall crispness, and implying that XviD is sharper because of this. This may very well be true, but there are still other factors involved, so you have to take everything into consideration--including implementation, not just features.

Anyway, if you can agree with me on the last statement, I think we can say we've reached a consensus. I'm going to do some tests with Divx vs. XviD, if I can only find a dev-api-4 build somewhere. I haven't really looked too hard yet, and I'm lazy. Know a good place to get it?

[ominte]

Originally posted by FreakNGoat
Though XviD has many more customizable features than Divx, does this mean that XviD is sharper? Maybe, maybe not. It could be possible that Divx just works better with the features it does have. Because XviD supports supposedly "better" quant types, does this mean it produces sharper output? Perhaps, but saying that it does exclusively because of quant type would be incorrect. Now it seems you have backed off of this argument slightly, to say that quant type plays a very important role on the overall crispness, and implying that XviD is sharper because of this. This may very well be true, but there are still other factors involved, so you have to take everything into consideration--including implementation, not just features.
Everything you are saying makes perfect sense but you seem to be dodging the bullet. In a practical situation if DivX 5 and XviD were to use the same features except XviD would use MPEG quantization then XviD would definitely produce the sharper result. Of course, using MPEG quants below a bitrate of at least 1200kb/s would produce unreasonable mosquito noise. Simply put there is no way around this, MPEG was designed, although not specifically, to produce a sharper result than H.263.

There is little need to worry about the developer's implementation because for both codecs the features being discussed are in reality fully implemented. Both H.263 and MPEG are international standards with development stopped long ago. So the question is not about the implementation on the developer side but on the side of the user. It is important for the user to determine when it is appropriate to utilise certain features as determined by bitrate.

[FreakNGoat]

Originally posted by ominte
Everything you are saying makes perfect sense but you seem to be dodging the bullet. In a practical situation if DivX 5 and XviD were to use the same features except XviD would use MPEG quantization then XviD would definitely produce the sharper result. Of course, using MPEG quants below a bitrate of at least 1200kb/s would produce unreasonable mosquito noise. Simply put there is no way around this, MPEG was designed, although not specifically, to produce a sharper result than H.263.


In a way I was dodging the bullet, but to prove a point. If you used MPEG quant and the rest of the codec implementation is crap, then you're probably not going to get sharper images.


There is little need to worry about the developer's implementation because for both codecs the features being discussed are in reality fully implemented. Both H.263 and MPEG are international standards with development stopped long ago. So the question is not about the implementation on the developer side but on the side of the user. It is important for the user to determine when it is appropriate to utilise certain features as determined by bitrate. [/B]

Just because they are fully implemented, does this mean they are implemented the exact same way? Its possible, I don't know enough about their inner workings to know. But look at MP3--Lame vs. Xing. Both produce output compliant to the MPEG specification, but the quality of their output is vastly different, because they use totally different algorithms and psych models. They both support the same features of the MPEG spec--VBR, joint stereo, etc... That was my whole point.

[Enigmax]

It will find a unloading of dev-api-4 compiled here:
http://www.divxhouse.com/
Remember, only recommended for tests, still are not ready for its publication.

Greetings

[mikeson]

@FreakNGoat:
If you used MPEG quant and the rest of the codec implementation is crap, then you're probably not going to get sharper images.
In theory, yes. But we are talking about DivX and XviD in their latest versions, so this is completely irrelevant.
Just because they are fully implemented, does this mean they are implemented the exact same way? Its possible, I don't know enough about their inner workings to know. But look at MP3--Lame vs. Xing. Both produce output compliant to the MPEG specification, but the quality of their output is vastly different, because they use totally different algorithms and psych models. They both support the same features of the MPEG spec--VBR, joint stereo, etc... That was my whole point.
How can you say that comparing DivX and XviD is similar to comparing LAME and Xing? :o You are comparing two things on completely different quality level. Have you done some test with DivX and XviD on your own? I doubt it, because you wouldn't talk like this. You should better start here (http://www.doom9.org/codecs-103-1.htm) and don't spread misinformations about bad implementation of MPEG4 features in codecs (and you haven't even prove it).

[temporance]

Originally posted by ominte
Everything you are saying makes perfect sense but you seem to be dodging the bullet. In a practical situation if DivX 5 and XviD were to use the same features except XviD would use MPEG quantization then XviD would definitely produce the sharper result. Of course, using MPEG quants below a bitrate of at least 1200kb/s would produce unreasonable mosquito noise. Simply put there is no way around this, MPEG was designed, although not specifically, to produce a sharper result than H.263.With respect, I think you are missing something here. MPEG-4 specifies exactly how the decoder works, how it should decode every feature (including MPEG quant). If I have two MPEG-4 encoders using the exact same features, there is nothing to say that those encoders do anything in the same way. The creators of these two encoders can make them do anything they want. provided the bitstream they generate is syntactically correct. Last time I looked at the xvid code, it was using the default forward quantiser code for both H263 and MPEG. There is no rule that says an MPEG-4 encoder must do this.

In short, what I'm saying is because xvid/MPEG is better than xvid/H263, then it does not follow that xvid/MPEG is better than divx/H263.

[FreakNGoat]

Originally posted by mikeson
In theory, yes. But we are talking about DivX and XviD in their latest versions, so this is completely irrelevant.


You are obviously completely missing my point. This whole argument "was" about theory.


How can you say that comparing DivX and XviD is similar to comparing LAME and Xing? :o You are comparing two things on completely different quality level. Have you done some test with DivX and XviD on your own? I doubt it, because you wouldn't talk like this. You should better start here (http://www.doom9.org/codecs-103-1.htm) and don't spread misinformations about bad implementation of MPEG4 features in codecs (and you haven't even prove it).

Are you serious? When did I ever say anything about "bad" implementations in codecs? I thought you wanted to avoid a flame war, but you've already talked down to me twice like I'm a complete troll. The LAME and Xing thing was simply an example to show how two things can be implemented differently, while producing output to a certain specification. You apparently read WAY too far into that one, perhaps thinking I was comparing LAME to Divx and Xing to XviD.

Yes, I have done my own tests. What does this have to do with anything I've been saying? Go back and read what I've wrote. I've seen that codec comparison, but the fact that you pointed me there obviously means you fail to understand what I have said. Never once did I claim Divx was sharper than XviD, which apparently you think I am doing.

You continuously keep missing my point, and seem intent on defending something you think I'm attacking. I'd say this argument has reached a level where it can go further, mainly based on the accusations you have placed against me which I don't appreciate and believe are totally unfounded.

[FreakNGoat]

Originally posted by temporance
In short, what I'm saying is because xvid/MPEG is better than xvid/H263, then it does not follow that xvid/MPEG is better than divx/H263.

Very well put. That is exactly my point as well.

[FreakNGoat]

Originally posted by Enigmax
It will find a unloading of dev-api-4 compiled here:
http://www.divxhouse.com/
Remember, only recommended for tests, still are not ready for its publication.

Greetings

Thanks Enigmax. I'll check that out.

[Enigmax]

Sorry, but the best codec now is the new VP6, from ON2 Technologies.
¡Fantastic!:D

Greetings

[SeeMoreDigital]

Originally posted by Enigmax
Sorry, but the best codec now is the new VP6, from ON2 Technologies.
¡Fantastic!:D Well it's good at bitrates below 700kbps. But it's GUI is not that user friendly at the moment and it's even slower than DivX to generate encodes - RV9 is much better though!

I'm certainly going to stick with DivX for a good while yet!

Cheers

[Enigmax]

He is faster than DivX 5.1.1 in Slowest mode and gives better quality (speed is similar to XviD). The GUI is very simple and of easy use.
I do not understand your problem?

Greetings

[Tommy Carrot]

Originally posted by Enigmax
Sorry, but the best codec now is the new VP6, from ON2 Technologies.
¡Fantastic!:D

Greetings

Sorry, but WRONG! :D

It's better in some cases, but the quality fluctates too much, and this codec has a tend to blur away the finer details. I dare to say xvid keeps more details, and the quality is more stable.

Of course, VP6 has the possibility, but it's not perfect yet (especially the rate-control).

[jggimi]

Gentlemen, please be civil. There is no "best." Each has it's attributes, positive and negative, and each will have its adherents.

@ Enigmax: if you were expecting flames, you got them, so you should be happy. If you weren't expecting them, you should have, making a general, sweeping claim like that in this forum. Many would consider it "baiting." And I'm one of those that do. Please avoid doing so again.

[mikeson]

Absolutely agreement with jggimi with no doubts. This way I also want to apologize to FreakNGoat for being a little bit rude, but I'm too allergic to statements like this is better than that. I think that we should share experiences instead of dogmas. As joggimi said, this doesn't belong to this board.

[Tuning]

I think doom9 is getting ready for next codec comparison.Lets wait untill its over.If XviD 1.0 had released it would have been better.

[Enigmax]

I feel it if annoying to somebody. But it is the comparison of Doom9 dogma?
Simply I think that codec is better than another at the most similar one to the original one is. Nothing else.
And I am loving of XviD. But VP6 has surprised to me pleasingly.

It is not my intention to bother to anybody. Sorry for my poor english.

Greetings

[Prettz]

Comparing Divx5 to Xvid with MPEG quant is like comparing apples and oranges. If you want a perfectly even comparison, only compare Divx5 to Xvid with H.263 quant.

Otherwise, there can't really be any comparison other than "do you like a smoothed look or a detailed/noisy look?"

As to comparing Divx5 to Xvid with H.263 quant, it's my opinion that Divx5 tends to over smooth the image far too much. There might be some scenes where both codecs will tend to over smooth, and some scenes where Divx5 over smooths but Xvid does not. So in that respect I would say that using Xvid is a safer bet for trying to preserve detail (once again, IMO).

[CruNcher]

Originally posted by ZZZERO
For those with lots of experience using Qpel, how does having GMC on or off affect the results?

i want to take on something i said in the XviD Section

Originally posted by CruNcher


for example


qpel+vhq4+cm+trellis+b-frames: PSNR 40.1454, SSIM 0.952205 ,8.107.008 bytes, 00:08:35.730
gmc+vhq4+cm+trellis+b-frames: PSNR 40.1558, SSIM 0.954120 ,8.105.472 bytes, 00:08:34.047
qpel+gmc+vhq4+cm+trellis+b-frames: PSNR 40.2188, SSIM 0.952830 ,8.106.496 bytes, 00:12:05.997


i dont agree here for me the qpel+gmc combination looks more detailed resambles the original the most the GMC only is more blurry details preservation is not as good

if you want to read the full story head over here http://forum.doom9.org/showthread.php?s=&threadid=63735