3ivx or xvid?

[boombastic]

I just bought a dvd-recorder which can also play avi.It should also play avi made with 3ivx.What are the most important features of this codec?Is there a reason for choosing it instead of xvid?Is it faster or slower startinf with the same avs file?

[bond]

to find out what the 3ivx codec supports have a look at the mpeg4 sticky

which codec you prefer depends on what you want to do, the best would be to test it yourself
but most here like xvid :)

[DeathTheSheep]

I guess "most" isn't me, especially for low bitrates... ;-)

[boombastic]

So do you think that at low bitrate 3ivx is better than xvid?

[DeathTheSheep]

Actually yes, that's what I think. XVid aims for crystal-clear quality at higher bitrates (I'm talkin' bitrates from 700kbps up). It can't even reach the lower end of the scale, for it is completely optimized for B-frames (which don't work well at these low bitrates).

I think 3ivx is indeed better at compressing video at lower bitrates (at lower quality, of course, but higher quality than Xvid at bitrates below 700kbps).

This is my $.02.

[Manao]

> for it is completely optimized for B-frames

That's plain wrong. Without bframes, Xvid is on par quality wise with DivX 6, while being 3 times faster.

[DeathTheSheep]

That's plain wrong. Without bframes, Xvid is on par quality wise with DivX 6, while being 3 times faster.

No, I'm afraid it isn't plain wrong, nor is it all too wrong at all for the bitrates I speak of.

As for on par with DivX6, that is true (perhaps a good bit better) at higher bitrates. However, for the bitrates I'm referring to (under 700 or even 500), 3ivx has even more speed and slightly better (versitile) use at low bitrates.

I speak of 3ivX5, of course, which can reach almost as low as DivX6 in terms of sheer compressability at low br.

[IgorC]

With Divx you can´t do 1 pass at fixed quant in 2 PASS encoding. Divx hasn´t more than 100 custom matrices (for anime , for films, low or high bitrate). Divx hasn´t advanced flexible options for 2pass (RC)as Curve Compression etc. Cartoon Mode. At all kind of bitrates Xvid advanced user ALWAYS will be enable to do better ASP coding.

And If you don´t know how to control all these features anyway you haven´t right to say that Xvid is worse. Just open Xvid forum and read read read read read and read it.

EDIT for clarity

[Manao]

DeathTheSheep : Sorry, but on the tests I made, 3ivx is far behind xvid even on high quantizers. It reaches in fact the quality of xvid vhq0 and it doesn't even have chroma motion ( which at low bitrates on animes is mandatory, because chroma artifacts are ugly at those bitrates )

boombastic : search the forum a little to understand what DeathTheSheep means by low bitrate ( hint : http://forum.doom9.org/showthread.php?s=&threadid=88268&highlight=coolness ).

edit : IgorC : i spoke of DivX because I consider it to deliver the best ( and slowest ) quality on a Pframe only encoding, whatever the quantizer. But the subject here is 3ivX vs XviD.

[DeathTheSheep]

With Divx you can´t do 1 pass at fixed quant.

Um, actually, you can in unrestrained mode. Check out the 1461 HiggsBason beta (or something like that... get it at www.divxlabs.com).

Chroma motion, eh? True, it does boost anime quality at low bitrates...

Heheh, yeah, I'm quite a BR fiend. :scared: :devil:

But later I used quant 31, which is darned good comparatively... So my test was indeed fairly accurate. But now B-frame deblocking has been implemented, it is no longer an issue, so thread closed. Yeah, the good old days.

[Manao]

> But later I used quant 31, which is darned good comparatively...

Comparatively to what ?

> But now B-frame deblocking has been implemented, it is no longer an issue, so thread closed

Huh ? What does it have to do with the topic ? we're talking about mpeg4 ASP codecs, and bframes for these codecs were _always_ deblocked. Do not confuse inloop deblocking and postprocessing deblocking.

[Sagittaire]

Originally posted by IgorC
With Divx you can´t do 1 pass at fixed quant in 2 PASS encoding.


With defaut setting, XviD quant don't use fixe quantizer in quant mode (bframe ratio/offset) like DivX ...



Divx hasn´t more than 100 custom matrices (for anime , for films, low or high bitrate).


H263 quantisation + PP4 is visually and with metric by far the best ASP quantisation mode for all source ...



Divx hasn´t advanced flexible options for 2pass (RC)as Curve Compression etc. Cartoon Mode. At all kind of bitrates Xvid advanced user ALWAYS will be enable to do better ASP coding.


For me the RC for DivX is better than XviD. DivX use Bitrate Modulation and VBV and it's good for me. Who understand really the advanced Rate Control setting for XviD like Overflow setting or difference between HighBitrateReduce or LowBitrateBoost ... !??

[Maximillius]

Originally posted by Sagittaire
With defaut setting, XviD quant don't use fixe quantizer in quant mode (bframe ratio/offset) like DivX ...

Then all you do is change the ratio to 1 and offset to 0. Not really that hard.

Originally posted by Sagittaire
H263 quantisation + PP4 is visually and with metric by far the best ASP quantisation mode for all source ...

If this were true, no one would be spending them time to make other matrices, right? :) Maybe for you this is true, but to my eyes H263 is not the best on all sources, a few maybe, but definitely not all.

Originally posted by Sagittaire
For me the RC for DivX is better than XviD. DivX use Bitrate Modulation and VBV and it's good for me. Who understand really the advanced Rate Control setting for XviD like Overflow setting or difference between HighBitrateReduce or LowBitrateBoost ... !??

With some reading about it, its not that hard to figure it out.

[IgorC]

In this moment the differnce beween Divx fusion (last beta )and Xvid (last Xvid 1.1 beta 1 march 14 )is too small to talk about it in low bitrates. But Xvid at high bitrate is better . Maybe OPSNR of Divx a litlle higher than of Xvid (visually high OPSNR is often just blurer video ;) ), but Xvid has a better SSIM result. And We don´t speak about the speed (another important factor).

Ok, it´s a little hard to tune all those PRO features of Xvid for the first time (Matrix, RC and other) but from technical point of view Xvid has mayor grade of choice what options to use. Maybe H263+ is better for OPSNR or SSIM , but not for my eyes. Custom matrix will always handle video in better way.

[Sagittaire]

Originally posted by Maximillius
Then all you do is change the ratio to 1 and offset to 0. Not really that hard.


with DivX too in reg (DivX use 150% for bframe ratio in reg) but for best quality you must use higher quant for bframe ...



If this were true, no one would be spending them time to make other matrices, right? :) Maybe for you this is true, but to my eyes H263 is not the best on all sources, a few maybe, but definitely not all.


Default setting is and will be always H263 quantisation for XviD because with PP the H263 matrix is the best and by far in all the situation. Without PP it's another problem ...



With some reading about it, its not that hard to figure it out.

Difference between HighBitrateReduce and LowBitrateBoost ... ???

[Maximillius]

Originally posted by Sagittaire
Default setting is and will be always H263 quantisation for XviD because with PP the H263 matrix is the best and by far in all the situation. Without PP it's another problem ...

If I want a soft looking picture in all cases then I guess H263 would be best on all sources. I don't like this look hence why it's not always best. I don't think I should have to oversharpen all my sources to cancel out H263's smoothing of fine details, and on HiDef encoding this smoothing is especially noticeable. Which is why if I do not use a custom matrix I always use MPEG because I think it looks far better in most cases then H263.

[Sagittaire]

Originally posted by IgorC
In this moment the differnce beween Divx fusion (last beta )and Xvid (last Xvid 1.1 beta 1 march 14 )is too small to talk about it in low bitrates. But Xvid at high bitrate is better . Maybe OPSNR of Divx a litlle higher than of Xvid (visually high OPSNR is often just blurer video ;) ), but Xvid has a better SSIM result. And We don´t speak about the speed (another important factor).

Ok, it´s a little hard to tune all those PRO features of Xvid for the first time (Matrix, RC and other) but from technical point of view Xvid has mayor grade of choice what options to use. Maybe H263+ is better for OPSNR or SSIM , but not for my eyes. Custom matrix will always handle video in better way.

In fact DivX6 beta seem very better for low quant. XviD is very better for SSIM with AQ (psy mode) and because psy mode for DivX seem not optimized at this time (decrease quality ... visually and with SSIM) ...

http://labs.divx.com/archives/TESTFusion.pdf

In fact DivX6 could be the best MPEG4 ASP codec ...

[IgorC]

Maybe. Let´s wait for a Divx6 and Xvid 1.1 final versions.

[Maximillius]

Originally posted by Sagittaire
In fact DivX6 could be the best MPEG4 ASP codec ...

Maybe if you only go by metrics, but I've never liked the look that DivX gives. Even at higher bitrates it smooths out far too much detail and at low bitrates, ugh, DivX looks extremely ugly. High metrics don't mean squat if you can't stand the look of the end result.

[Sagittaire]

DivX 5.2.1 and DivX Fusion (DivX6 beta) aren't the same codec. DivX Fusion is very better than DivX 5.2.1.

High metrics don't mean squat if you can't stand the look of the end result.

IMO you don't test DivX Fusion: High metrics don't mean squat if you can't stand the look of the end result but probaly that make no test and make conclusion are more ridiculous ... lol

[Maximillius]

Originally posted by Sagittaire
DivX 5.2.1 and DivX Fusion (DivX6 beta) aren't the same codec. DivX Fusion is very better than DivX 5.2.1.

I never make that claim they are the same. I've seen the results of Divx Fusion and I don't like the look of it. End of story.

Originally posted by Sagittaire
IMO you don't test DivX Fusion: High metrics don't mean squat if you can't stand the look of the end result but probaly that make no test and make conclusion are more ridiculous ... lol

And you make the claim that I haven't done any tests from what evidence? Making erroneous assumptions about me as you have done is quite ridiculous.

[IgorC]

I discovered myself that PSY of Divx6 Fusion beta2 doesn´t work good and 3 weeks later Sagitarrie announced a big secret :o that PSY parametlicly isn´t good. We already know it. Divx Team also said that
PSY doesn´t work good yet.
I also don´t like how Divx6 is looking. The video is so washed. And We don´t care about OPSNR, SSIM at all.

[Maximillius]

Originally posted by IgorC
I also don´t like how Divx6 is looking. The video is so washed. And We don´t care about OPSNR, SSIM at all.

Yep, what's pleasing to the eye is far more important to me then what some metric says.

[temporance]

But what is doing the washing? Is it the encoder, or the postprocessing? IMHO, the DivX Fusion deblocking is the best for metrics, but maybe some people don't like the look of it.

On the encode side, I'm with Sagittaire, DivX has the edge, especially at high bitrates.

[Maximillius]

Here's an example of what I'm talking about:

Huffyuv

http://img207.exs.cx/img207/9615/source6bs.th.png (http://img207.exs.cx/my.php?loc=img207&image=source6bs.png)

XviD (Overall PSNR: 43.3417, PSNR of frame: 46.4234)

http://img207.exs.cx/img207/3536/xvid4pe.th.png (http://img207.exs.cx/my.php?loc=img207&image=xvid4pe.png)

DivX Fusion (Overall PSNR: 42.8322, PSNR of frame: 43.6493)

http://img207.exs.cx/img207/3633/divx1jg.th.png (http://img207.exs.cx/my.php?loc=img207&image=divx1jg.png)

Huffyuv all defaults.

XviD settings: Motion Search: 6; VHQ 4; VHQ for b-frames; chroma motion; chroma optomizer; b-frames: 2 consecutive, ratio:1, offset:1.

DivX settings: Unconstrained; Insane Quality, 1 consecutive b-frame; no psy enhancements.

Both were encoded single pass at quant 6.

Avisynth script used:

mpeg2source("thing.d2v",iPP=true,cpu=4)
tfm(d2v="thing.d2v",pp=0)
tdecimate(mode=1)
BlindDeHalo3(rx=2.5,ry=2.5,strength=100,lodamp=0.0,hidamp=0.0,tweaker=0.0,PPmode=1)
LimitedSharpen(ss_x=1.75,ss_y=1.75,Smode=3,strength=150,wide=true)
DeGrainMedian()
Trim(30000,30050)

As you can clearly note in this shot that DivX is smoothing way beyond what is necessary. You can notice in the jacket where some of the lines are being smudged out. PSNR values are gotten from Avisynth using compare function (only given since I know you like metrics Sagittaire). Sorry, but I don't see anything spectacular about DivX Fusion and this smudging of detail is very noticeable especially in more high detail scenes then this and is very ugly, but I believe this screenshot shows what I mean pretty well.

[stephanV]

both screenshots (DivX/XviD) are equally bad. The blocks in XviD are much more clear. Using ratio 1 for b-frames kinda defies the purpose of them and it wouldnt surprise me if the XviD encode was much larger as the DivX encode.

[Maximillius]

Originally posted by stephanV
both screenshots (DivX/XviD) are equally bad. The blocks in XviD are much more clear. Using ratio 1 for b-frames kinda defies the purpose of them and it wouldnt surprise me if the XviD encode was much larger as the DivX encode.

Unless you consider 15KB "much larger" then no. I can redo it changing the ratio, but it doesn't change the fact that DivX is smoothing the picture much more. That was my whole point.

[stephanV]

on 50 frames that could make a difference... what are the 2 bitrates? And XviD blocks more... so whats better?

:)

[Maximillius]

It's around 710kbit for XviD and 650kbit for DivX. I reencoded it with the 2.00 for ratio and now the XviD 635kbit. New overall PSNR = 43.0718 and PSNR for the frame is still 46.4234 so it didn't really change much and is still higher then divx.

Here's a new screenshot of the same frame:

http://img207.exs.cx/img207/511/xvidnew5zd.th.png (http://img207.exs.cx/my.php?loc=img207&image=xvidnew5zd.png)

Comparing this new one to the DivX one you can still see that Divx is smudging the lines on the jacket. As is always the case between DivX vs. XviD, XviD shows more detail but a bit more blocks, DivX has less blocks but smudges more. I prefer the XviD look. That was the point of my previous comments. :)

[Maximillius]

Originally posted by stephanV
And XviD blocks more... so whats better?

I see a bit of blocking, but not much at all. During playback these blocks aren't noticeable, but the smudging is. So I'd say XviD is better.

[Maximillius]

Here's with the contrast increased to show off the jacket more which is where I guess you were referring to.

XviD:

http://img207.exs.cx/img207/4394/xvidcontrast1wc.th.jpg (http://img207.exs.cx/my.php?loc=img207&image=xvidcontrast1wc.jpg)

DivX:

http://img207.exs.cx/img207/9109/divxcontrast6bn.th.jpg (http://img207.exs.cx/my.php?loc=img207&image=divxcontrast6bn.jpg)

There isn't much different between the two with the blocking, but that nots really the point of this comparison which for both isn't that noticeable during playback (since the PP takes care of most of it). What's really important is how much more DivX is smoothing which is very ugly during playback. I think I've said enough. :)

[temporance]

Max: Could what you're seeing be an effect of postprocessing?

[Maximillius]

Originally posted by temporance
Max: Could what you're seeing be an effect of postprocessing?

There is no post processing going on. I loaded both encoded avis in virtualdub and had it take screenshots. Both decoders are set to no PP anyway.

[IgorC]

It´s not deal of PostProc (PP). I also try differnets PP with differnets decoders (Divx, Xvid, FFDSHOW). The same result.

Best SSIM for Divx is decoding by Divx PP4 (without deringing). Deringing is just wash it more and picture is still uglier

[Maximillius]

Originally posted by temporance
On the encode side, I'm with Sagittaire, DivX has the edge, especially at high bitrates.

At higher bitrates DivX does smooth less, but it still smooths more then I would like. XviD still looks better to my eyes at higher bitrate for a sharper picture. But then again this all comes down to personal preference and no metric either way is gonna be more important then your own eyes.

[Sagittaire]

XviD (Overall PSNR: 43.3417, PSNR of frame: 46.4234)
DivX Fusion (Overall PSNR: 42.8322, PSNR of frame: 43.6493)

I fact here compare are very useless ... for me with 46.4234 vs 43.6493 it's really useless to make visual comparison: 46's frame is always very better than 43's frame (perhabs bframe for DivX and Pframe for XviD)

If you want make really compare XviD vs DivX:

1) Size file must be identical particulary for high quant encoding. Encoding with less or more +/- 1 Kbps for example.

2) Use same setting for XviD and DivX:
- bframe 3/1.50/0.00 for XviD (it's the default setting for multibframes DivX)
- H263 + trelli for XviD and H263 optimized for DivX
- No psy, No GMC, No Qpel

[Maximillius]

Originally posted by Sagittaire
- bframe 3/1.50/0.00 for XviD (it's the default setting for divX)

Default is 1 consecutive b-frame for DivX not 3...

[Sagittaire]

multibframe for DivX : 3/1.50/0.00 for maxbframe/ratio/offset

quantizer mode for DivX and XviD are not equivalent: XviD use less size than DivX for same quant ...

[Maximillius]

Originally posted by Sagittaire
multibframe for DivX : 3/1.50/0.00 for maxbframe/ratio/offset

Well it's default to 1, which is what I used. XviD had more b-frames at a higher quant, so I'm not sure what you complaint is.

[Maximillius]

Also after I changed the ratio back to 2.00, the XviD had a 15kbit lower bitrate and 1 more consecutive b-frame at higher quants then the divx encode. So basically DivX had the long end of the stick and still didn't impress me. Basically let me say it now, no matter how you try to get me to fix it even more for DivX I'm still not gonna like it. You're just gonna have to live with it.

[IgorC]

Why should we use H263 for Xvid? Who is obligating me?

[Maximillius]

Here are stats using DRF analyzer just to prove that I didn't stack anything in XviD's favor:

XviD:

http://img187.exs.cx/img187/2526/drfxvid16er.png
http://img187.exs.cx/img187/125/drfxvid26xz.png

DivX:

http://img187.exs.cx/img187/9129/drfdivx19sm.png
http://img187.exs.cx/img187/4466/drfdivx29oq.png

DivX has a lower average quant and less b-frames and yet it still looks worse and has lower PSNR. I believe I was more then fair to DivX and it failed to impress.

[IgorC]

Don´t forget we are talking about visual quality here. Not metric ;)

[Sagittaire]

@ IgorC

In fact H263 + PP4 is IMO visually and by far very better than CM + PP0 or CM + PP4 : more blocking but less ringing for H263 in PP0 but with PP4 H263 blocking problem is solve.

If you use only PPO it's another problem. HVS CM could be visually better with PPO : more ringing but less blocking for HVS CM in PP0.

[Maximillius]

Originally posted by Sagittaire
In fact H263 + PP4 is IMO visually and by far very better than CM + PP0 or CM + PP4 : more blocking but less ringing for H263 in PP0 but with PP4 H263 blocking problem is solve.

Again this is just your opinion which some people don't share as I don't like the overly smoothed look it gives. Let me just say this to end it, its great that you like DivX and I have no problem with that. But I don't like it and that's just the way it is. I'm done and have a good day. :) I'm off to play CS source.

[Sagittaire]

@ Maximillius

In fact your source is probally low motion source (use high max bframe number and high PSNR for Q6). In this case MaxBframeNumber is very important ... use multiple bframe for DivX is very important.

H263 is the best quantisation method for MPEG4 ASP and by far: XviD, DivX, NDASP and all ASP codec use and will use always H263 for default setting. Speack about dev like Digital56K or XviD dev if you want ...

[IgorC]

Sagitarrie, you have your opinion about best matrix and other people have another opinion.
Why we don´t try to do visual test , not metric . Because finally visual quality is what we all want (at least mayor of us). Screenshots is not a good idea, but some samples (not optimized for metric quality but for visual)

If H263 is best matrix , why there is test of matrices in Xvid forum?

[Maximillius]

Well since I'm still waiting for a space to open up on my CS server I'll respond again.

Originally posted by Sagittaire
In fact your source is probally low motion source (use high max bframe number and high PSNR for Q6). In this case MaxBframeNumber is very important ... use multiple bframe for DivX is very important.

Akira is far from being a low motion source. And in the more high motion/high detail scenes DivX smoothes things out to a point that I might as well be watching a low resolution VHS tape rather then a DVD rip.

Originally posted by Sagittaire
H263 is the best quantisation method for MPEG4 ASP and by far: XviD, DivX, NDASP and all ASP codec use and will use always H263 for default setting. Speack about dev like Digital56K or XviD dev if you want ...

Again you're speaking of opinions as fact. H263 is not the best quantization method for ASP unless you like your sources to be oversmoothed which I don't. And I could honestly care less what a dev tells me. I go by my eyes, not what a metric or someone else tells me. Secondly, if this were honestly the case why would there even be an option to have custom matrices? If H263 is honestly the best matrix ever then this option would be completely worthless, no? But apparently not ever loves H263 as much as you which is why they come up with custom matrices that perform better to what they want. I prefer a matrix that preserves more details rather then smoothing them out even if this leads to a bit more blocks. You may not prefer this and that's fine. But please, don't speak of your opinions as if they are solid facts that everyone agrees on, because they aren't.

Since we have significantly strayed from the OT so much, this will be my last post in the thread. Feel free to continue this without me.

[temporance]

Originally posted by IgorC
If H263 is best matrix , why there is test of matrices in Xvid forum? IIRC, they did double-blind testing of clips encoded with various parameters and one encoded using H.263. The H.263 encoded clip was voted the best looking: the result was very clear.

[Sagittaire]

@ Maximillius

lol ... you don't encode complete movie but only "low motion" 50 frames: I repeat but 43 dB for Q6 encoding is typical "low motion scene"

For example generaly HVS matrix are more agressive than H263 for High Frequencies (detail level is worst -> more ringing) but less agressive for low Frequencies (gradiant is good -> less blocking)


@ temporance

Yes, H263 is always in the first visually (see Soulhunter visual comparison matrix) and it's probaly with PP0. With PP4 H263 is IMO and by far the best ...