DiVX 4/5 comparison test

[sarahjh69]

I just got DiVX 5 and have done a few comparisons
all tests done using cell 2 of armageddon and athlon xp1600
XMpeg 4.02 2 pass encoding with 1403 selected as bitrate
full size crop top and bottom and 72 off each side (xmpeg crop)

slowest setting
divx 4............22-26 fps....53 meg file
divx 5............28-32 fps....54 meg file......25% faster than 4
divx 5 +3 options 14-28 fps....52 meg file......20% slower than 4

I don't think the 3 options have made enough diff
to file size or quality to make it worth slowing down
so a 25% speed increase on divx 5

medium setting
divx 4............26-32 fps....53 meg file
divx 5............28-34 fps....54 meg file.......10% faster than 4

between all the encodings there was a very small difference in quality between slowest and medium settings but not between
different versions of divx. also on divx 5 pass 1 would be at the
slower fps and pass 2 would be around the higher fps.

Conclusion
So it seems I can encode at slowest on divx 5 at the same fps
as medium on divx 4! (or stick to medium and do it 10% faster)
I am not going to use the 3 extra encoding options at this time,
far too slow and no obvious benefit!

[Difflam]

> I don't think the 3 options have made enough diff
> to file size or quality to make it worth slowing down
> so a 25% speed increase on divx 5

Those 3 extra options are there to increase quality, not decrease file size... so if you are going for max quality, you better use em! :)

> I am not going to use the 3 extra encoding options at this time,
> far too slow and no obvious benefit!

Like I said... they increase quality. Its possible that the benefit gained isnt very noticable at such high bitrates... but I'm guessing they would make a difference at lower bitrates (sub 1000).

[khp]

Since it's imposible to judge quality objectivly, I have done the following the test in constant quality mode.

All the encodings were done without audio.
The Divx5.0 basic test was done with both the standard and the pro version with identical results.

Divx4.12 filesize 1.48GB (1,590,431,744 bytes)
Divx5.0 Basic 1.47GB (1.580,339,200 bytes)
Divx5.0 Q,G 1.34GB (1,446,283,264 bytes)
Divx5.0 B,Q,G 1.04GB (1,124,216,832 bytes)

B = using B-frames.
Q = using Quarter pel motion compensation.
G = using Global motion compensation

The filesize reduction gained from using B frames is quite impressive, if the quality is indeed unchanged.
Also worth noticeing that the basic version of divx5 offers practically no improvement over divx4.12, I am kinda dissapointed that DXN would release something like that an call it Divx5, the pro version is worthy of the name, but not the standard version. It's kind of like calling the Geforce4mx a Geforce4.

Comments are very welcome.

[Darrius "Junto" Thompson]

Originally posted by khp
Since it's imposible to judge quality objectivly, I have done the following the test in constant quality mode.

All the encodings were done without audio.
The Divx5.0 basic test was done with both the standard and the pro version with identical results.

Divx4.12 filesize 1.48GB (1,590,431,744 bytes)
Divx5.0 Basic 1.47GB (1.580,339,200 bytes)
Divx5.0 Q,G 1.34GB (1,446,283,264 bytes)
Divx5.0 B,Q,G 1.04GB (1,124,216,832 bytes)

B = using B-frames.
Q = using Quarter pel motion compensation.
G = using Global motion compensation

The filesize reduction gained from using B frames is quite impressive, if the quality is indeed unchanged.
Also worth noticeing that the basic version of divx5 offers practically no improvement over divx4.12, I am kinda dissapointed that DXN would release something like that an call it Divx5, the pro version is worthy of the name, but not the standard version. It's kind of like calling the Geforce4mx a Geforce4.

Comments are very welcome.

In our prelminary testing we've had similar results. If you have Spykids, give it a shot. For some reason it give incredible results with Qpel, B, and G. Not sure why. Analyzing why.

In regards to 5.0 Standard. I agree. The problem is we had Psychovisual Modeling working fairly well about a week ago until we broke it at the last minute. Sorry :( But Psychovisual Modeling certainly has a ton of potential, we're just now building some serious testing tools to dynamically see exactly how the modeling affects each block. I think some great things will happen here. However you are correct, it currently isn't to worthy of the 5.0 name. I encoded about 100 clips or movies during our dev cycle and can post the Min/Max/Avegare results of each feature if anyone is interested.

Darrius "Junto" Thompson

[khp]

Thank you very much for yor response, it's nice to know that DNX cares enough about our opinion, to keep an eye on what we have to say in this forum :) .
Or maybe you just came here because your own forums are awash in a flamewar about the price of the pro version (I paid the 30$).

Personally I don't care much for the Psychovisual Modeling, I don't think I will ever trust a machine to make that kind of alterations.

I encoded about 100 clips or movies during our dev cycle and can post the Min/Max/Avegare results of each feature if anyone is interested.

Very much interested.

[Teegedeck]

Could you try

Divx5.0 B,G
?

I'm very curious. ;)

[khp]

As requested:

Divx5.0 B,G 1.10 GB (1,190,881,280 bytes)

[Teegedeck]

Thanks a lot, khp! So quarterpel seems to actually reduce filesize as it should, in some cases. There were reports that it did the opposite. It should prove interesting to find some regularities here.

[Zarxrax]

I did a test on a clip about 1 minute long that was about half low-motion and half VERY high motion. The quartar-pel did indeed increase the file size by from about 16mb to about 17mb.

[sarahjh69]

I just found Xmpeg 4.31b
Its added 1 fps to the speed

so at medium with none of the extra options 30-33 fps
compared to divx 4.12 and Xmpeg 4.2a giving 26-32 fps

quite a good speed increase!

[FSB-SPY]

Originally posted by khp
Divx4.12 filesize 1.48GB (1,590,431,744 bytes)
Divx5.0 Basic 1.47GB (1.580,339,200 bytes)
Divx5.0 Q,G 1.34GB (1,446,283,264 bytes)
Divx5.0 B,Q,G 1.04GB (1,124,216,832 bytes)


Did those last two runs with the DivX5 codec prove to be less in terms of quality? Or is quality the same, just smaller filesize?

[robUx4]

My personal tests are here : http://mukoli.free.fr/video/

The interresting parts are the screen shots. There is a comparison between DivX5, DivX4 and XviD.

I put my conclusions, you have have yours :)

[khp]

@FBI-spy:

I can't tell much difference, but I didn't look very hard, and I don't consider myself to be a great judge of quality, so I don't feel comfortable telling one way or another, but I haven't noticed any artifacts in any of the encodings.
Anyway My test was primarily aimed at measuring the preformance of the MPEG4 options. Any quality issues will be a lot harder to judge an will require much more testing.

@robUx4:

Very interresting article, this is the kind of stuff we need, to make our choice between the different versions of divx.
The discoloration in some of the divx5 shots are obviously a serious problem, Hopefully the DARC team is busy hunting this thing down.

[khp]

@robUx4

About your problems with fast recompress, are you using a hacked version of divx5 pro ?, so far all my encodings has been done with fast recompress, and I have had no such problems.

[-h]

The interresting parts are the screen shots. There is a comparison between DivX5, DivX4 and XviD.

The "bitrate" tests aren't actually achieving what you want them to - a much more accurate measurement of codec quality would be constant quantizer encodes. This can be achieved by setting min quant = max quant in DivX4/5, and using the "1-pass - quantizer" mode in XviD. Such a test will allow you to be pretty sure that the image quality will be the same in all files (the same quantizer is being applied, after all), so you can then examine how large a file a given codec requires to reach a certain level of quality.

Also, the blocks you see in XviD are due to post-processing being absent. If you turned off post-processing for the DivX clips, they'd most likely have the same blocks :) Change the FourCC of the XviD clips to divx/DIVX, and you'll see the difference.

To sum up what you'll most likely find from a fixed quantizer test:

- XviD performs better than DivX4 in all circumstances.
- XviD is similarly better than DivX5, unless you use B-frames.

-h

[khp]

Originally posted by -h

The "bitrate" tests aren't actually achieving what you want them to - a much more accurate measurement of codec quality would be constant quantizer encodes. This can be achieved by setting min quant = max quant in DivX4/5, and using the "1-pass - quantizer" mode in XviD.

I don't agree, robUx4 does provide valid arguments for his testing methods (in my opinion), testing using fixed quantizers is good for initial testing, but in the end it all comes down to quality. It is impossible to judge quality objectivly, but testing with fixed quantizers doesn't necessarily tell the whole story.

[robUx4]

Originally posted by khp
@robUx4

About your problems with fast recompress, are you using a hacked version of divx5 pro ?, so far all my encodings has been done with fast recompress, and I have had no such problems.

Nope, I don't think so. I got it from a regular website that mirror some files.

[robUx4]

Originally posted by -h
The interresting parts are the screen shots. There is a comparison between DivX5, DivX4 and XviD.

The "bitrate" tests aren't actually achieving what you want them to - a much more accurate measurement of codec quality would be constant quantizer encodes. This can be achieved by setting min quant = max quant in DivX4/5, and using the "1-pass - quantizer" mode in XviD. Such a test will allow you to be pretty sure that the image quality will be the same in all files (the same quantizer is being applied, after all), so you can then examine how large a file a given codec requires to reach a certain level of quality.

Also, the blocks you see in XviD are due to post-processing being absent. If you turned off post-processing for the DivX clips, they'd most likely have the same blocks :) Change the FourCC of the XviD clips to divx/DIVX, and you'll see the difference.


That's not fair. I want to use a regular method. If the post processing is not available on XviD I'm not going to change the produced files.

Also my aim is not to test the absolute quality of the codecs, but to try to find the best way to archive the things I record from TV. Using fixed quality may be nice, but if it's too big to put on a CD, that's not very usefull for me.

I just say XviD is not good for my kind of use, maybe it is in other cases.

[-h]

I don't agree, robUx4 does provide valid arguments for his testing methods (in my opinion), testing using fixed quantizers is good for initial testing, but in the end it all comes down to quality. It is impossible to judge quality objectivly, but testing with fixed quantizers doesn't necessarily tell the whole story.

Exactly - we have an objective test of quality, and that test is encoding with a fixed quantizer. That will tell you, empirically, which codec is "better" than another. After you know that, all the quality of a rip depends on is how far you decide to (or are able to) tweak an encoding session.

All I was responding to was:

As everything was encoded in 1-pass with a bitrate respect constraint, it's important to see how the codec respect what you asked them. I chose the 1-pass mode because that's the one I use (more convenient) and it also gives a "pure performance" aspect of the codec. I consider that if a codec is bad in 1-pass it's going to have to do more complicated work in 2-pass mode.

The "pure performance" of a codec isn't its CBR performance, as that's just a test of its rate control system. The pure performance of a {motion estimation -> DCT} codec such as MPEG4, is how small a file is for a given quantizer and the time taken to encode said file.

-h

[-h]

That's not fair. I want to use a regular method. If the post processing is not available on XviD I'm not going to change the produced files.

XviD has had postprocessing in its dshow filter for a couple of weeks now - I assumed you had an old build (which wouldn't have post-processing available), or didn't turn it on in the filter properties.

A quick way to avoid downloading a newer XviD build was just to change the FourCC of the XviD clip, that's all. Or, when you encode with XviD, set the "FourCC used" option to DIVX, instead of XVID.

-h

[robUx4]

Also my main concern apart from the quality itself (which can be tweaked) is the motion problem I describe at the end. Download the 2 AVI (one DivX5 and one XviD) and see the problem.

I used the Very-High motion parameter.

[FSB-SPY]

Originally posted by khp
@FBI-spy:

I can't tell much difference, but I didn't look very hard, and I don't consider myself to be a great judge of quality, so I don't feel comfortable telling one way or another, but I haven't noticed any artifacts in any of the encodings.
Anyway My test was primarily aimed at measuring the preformance of the MPEG4 options. Any quality issues will be a lot harder to judge an will require much more testing.



I don't suppose you still have those AVIs? Could you perhaps take a few snapshots (or even little clips) comparing the DivX 4 and the 5 with those 3 extra settings? I'm currently in the middle of re-encoding some of my movies, and I was thinking of using DivX3.11 like others have been suggesting for the longest of time. But if DivX5 is better in terms of quality/file size [for said quality], then I'll move on to that instead.

XviD doesn't look so hot. But that's now I suppose, give it time and it may surpass. I don't have the time to give it though. ;)

[khp]

Originally posted by FSB-SPY

I don't suppose you still have those AVIs?

No, sorry they have been deleted, but I'am still testing, I'll post snapshots of artifacts, if I find them.

But given the amount of postings speaking of artifacts, even without any of the 3 mpeg4 options, I would not use divx 5.0 for anything bound for permanent storage.

So far we have only seen the first revision of divx5.0, and betatesting appear to have been minimal, let's wait for a new revision or two and give that a week or two of intensive testing.

ATM nandub may well be the your best option.

[-h]

Also my main concern apart from the quality itself (which can be tweaked) is the motion problem I describe at the end. Download the 2 AVI (one DivX5 and one XviD) and see the problem.

I downloaded the clips - the difference is post-processing being enabled for DivX5, but not for XviD. When post-processing is enabled for XviD, the blocks are far less obvious, and it looks very much like the DivX sample.

Also, because CBR mode (rate control) is being used, the codecs are giving unequal attention to different areas of the movie. In the clips from your site, the high-motion jeep scene looks noticably worse in the DivX5 clip, because its CBR system didn't give it as many bits as XviD's CBR system did. But since XviD gave those bits to the jeep scene, it had fewer for the scene change immediately following it, whereas DivX still had bits in reserve because it didn't spend them on the jeep.

So, we have a comparison of (untweaked) CBR engines, but not a comparison of the codec cores themselves (motion estimation, etc.). This is what you were after though, it seems, as all your encodes are CBR, and DivX5 appears to suit your needs well. I have just been noting that you can't draw conclusions of the codec's core performance, or 2-pass performance, or any other variable (excluding perhaps speed) from this, as CBR encoding introduces too much variation for objectivity.

-h

[robUx4]

Originally posted by -h
I downloaded the clips - the difference is post-processing being enabled for DivX5, but not for XviD. When post-processing is enabled for XviD, the blocks are far less obvious, and it looks very much like the DivX sample.


I love the way you can emphase the facts ;)

I've downloaded the lastest XviD binary from Keopi's site and made the test on other stuff. And still with post processing enabled (I've verified and tried also with the FourCC too) the problem is still there and much more noticeable than in DivX. Bu the overall image quality is better than DivX.

If this motion problem was fixed, I would go for the XviD... But I can't afford 2-pass most of the time (for example I have 3 things to record this evening).

[-h]

If this motion problem was fixed, I would go for the XviD... But I can't afford 2-pass most of the time (for example I have 3 things to record this evening).

Well, the latest (even "more latest" than Koepi's) build from CVS has Isibaar's completely rewritten CBR system. I don't know when Koepi will be uploading it though, and I'm in the middle of redesigning the gui so it might be best to wait..

Different is always good though. It might fix your problem, or make it worse. But at least it's progress :)

-h

[MaTTeR]

@robUx4

I didn't see much difference between the Xvid and Divx 5 clips either. Then again, most of the comparison tests I've seen in this forum is showing me the same:) Maybe I'm getting old.

-h is right though. The difference between the clips during playback are most likely post-processing. The DS for Divx 5 seems to blur or smooth the image thus giving you an effect that it's playing smoother or more seamless. I have to say I've never used a CBR mode with any codec and I always use 2 pass modes during comparison. I use 2ps for obvious reasons and I'm always dealing with DVD source. So this explains why I don't have a need for CBR.

IMO a 2ps comparison test will clearly show you the quality difference between codecs:) Need I say which image looks cleaner and holds more detail?:D Also notice which image holds closer to the true color of the source.

[robUx4]

Yeah it's true that with a DVD source you can "afford" to use 2 pass because the source is nice. In my case the TV is not so great. And even the cable operator compress too much the programs :( (digital drawback).

[AM Tiainen]

robUx4 test ( http://mukoli.free.fr/video/ ) was pretty good to show difference between codecs and DivX 5 options.

I have also own codecs comparisor. That found from http://www.nixed-hq.com and press "Codec Comparisor" link
or directly http://koti.mbnet.fi/website/mpeg4test/

[-h]

I have also own codecs comparisor. That found from http://www.nixed-hq.com and press "Codec Comparisor" link
or directly http://koti.mbnet.fi/website/mpeg4test/

Many of the captions say "DivX5 contained more details by far" or "DivX5 was clearly superior", although the screenshots look the same (?). For example frame 26376 for XviD and DivX5 - where are all the extra details that DivX5 has? Fine details, like the city out the window, are preserved in both. Same with 108872.

Also, I can't easily see the "added details" of SBC in 153906 when compared to the others.

When talking about sharpness, could you perhaps outline or mention a specific area or feature where the details are?

-h

[khp]

All in all a great article.

Divx3.11 does appear to suffer when compared to the others, but I must agree with -h's conclusions and add that I would probably fail to pickout the divx5 images, in a blind test. And your conclusions does seem a bit presupposed.

[AM Tiainen]

Originally posted by -h

Many of the captions say "DivX5 contained more details by far" or "DivX5 was clearly superior", although the screenshots look the same (?). For example frame 26376 for XviD and DivX5 - where are all the extra details that DivX5 has? Fine details, like the city out the window, are preserved in both. Same with 108872.

Also, I can't easily see the "added details" of SBC in 153906 when compared to the others.

When talking about sharpness, could you perhaps outline or mention a specific area or feature where the details are?

-h

We know that problem is that resolution is only 576 x 224. We want good result with every codecs so we take this resolution on movie. You can try to zoom with right mouseclick or use lower resolution to see easily difference. But if you looks video on full screen you notice easily difference. But when you see 1:1 image resolution all pictures looks almost same. So we had to analyze picture very closely.

Frame 26376 DivX 5 is better from cupboard (is look more closely to DVD material than XviD) and also plant on left bottom corner is sharpen than XviD. Click one time left mouse button to activate windows and use to keyboard number 3 & 4. Can you see difference?

Frame: 108872 You are right difference are in that picture minimal.

Freme: 153906 SBC is best of all. It is much detail when you watch around from right eye.