XviD review - in pictures

[pieroxy]

I did put up a review of encoding 'Finding Nemo' in XviD. I did several encodes: 500, 750, 1000, 1500, 2000, 3000, 4000kbps. Results are quite surprising - in terms of quality and target bitrate - and I sure didn't expect that out of the review...



Well, see for yourself:
http://nerds.palmdrive.net/video/21Feb2006/index.html

[Teegedeck]

When I read some sentences in the conclusion of this review ("'quantisation factor' = the factor by which the codec judge the quality of a frame ", "Where are all these bits going?") I think some mysteries would solve themselves for the author if he researched the relation between 'bitrate' and 'quantizer' and also searched for the term 'saturation'.

[ChronoCross]

that almost looks similar(the site layout and the actual method of comparrisson) to something I saw a few months back. and it came to the same dumbass conclusion.

http://nerds.palmdrive.net/video/Jan2006/index.html

Not only does it use outdated technology, there is no mention as to the xvid settings used. It looks like it goes directly from default settings in two pass mode.

Not the best (quality and time-wise) but that's what I used...

what the hell kind of backup person does that. jesus this test is automatically invalid.

Try using some of the advanced settings rather than always using out of the box settings, hell if your gonna use out of the box settings then a reference encoder will do pretty much the same.

SO PFFT to your test. I could make those look way better than you did simply by updating the filterset.

Mods: If you feel I'm out of line let me know and feel free to automatically edit my post.

[foxyshadis]

Default settings aren't bad settings, they're just a datapoint people on this forum don't care about, and others shouldn't need to. Really, the point where xvid's defaults were optimal was for the hardware of 2-3 years ago, though, and current average harware can do hardcore xvid in realtime or better for standard def. (I like the idea of including 6of9 and the various EQM matrices, but I wager they'd be abused or at least lead to many more "what's the best matrix".)

And yeah, 4000 kbps for cgi? C'mon now, and even mpeg2 could handle that. Complete lack of mpeg2 artifact removal, and a methodology of testing only screenshots (see doom9's comparisons) pretty much relegates this to the "pointless" bin though.

[sysKin]

I like this test a lot. In particular, the nice zoom thingy is EXCELLENT. What can I do to help Doom9 use that for codec comparison? Current Doom9's method (click on codec name to load this codec's result) is much worse for me (strongly IMHO).

what the hell kind of backup person does that. jesus this test is automatically invalid.
I disagree because this is a "constant bad thing" across all tests. It would only matter if he used something else for other encodings, and compared that.

I'd say the greatest flaw of this encoding is resolution - you should always resize from original, even just by 16 pixels (704x), to get rid of original mpeg2's blocks (or at least make them not align with new XviD's blocks).

[Teegedeck]

Default settings aren't bad settings, they're just a datapoint people on this forum don't care about, and others shouldn't need to. Really, the point where xvid's defaults were optimal was for the hardware of 2-3 years ago, though, and current average harware can do hardcore xvid in realtime or better for standard def. (I like the idea of including 6of9 and the various EQM matrices, but I wager they'd be abused or at least lead to many more "what's the best matrix".)Going a bit further off-topic for a moment, it would really be good to start working on some more presets, now that it is possible to include such things in the VfW config.

Having more presets some day is kind of unavoidable - because they represent a good thing and are possible now - but I fear arriving at them will not be easy. An anime or CG preset should be easy to do though - right?

How about a new thread 'Presets setup discussion'? sysKin, could you easily include more presets in the VfW config? Can such a preset encompass all kinds of switches, like even the quantizer for first pass, or are the possibilites limited?

[sysKin]

sysKin, could you easily include more presets in the VfW config?

I suppose so.
What I dream about, though, is making defaults, and presets, independant from VfW. Some kind of "xvid config" library that's shared by VfW, xvid_encraw, and all olther GPLed software that feels like it.

[Teegedeck]

That would be even better. :) So, as for limitations, I take it there are none? (Sorry, I might have edited my message even when you were writing your reply.)

Edit:And again, I think of the most important question only later: could a preset prompt loading an external matrix or would the matrix have to be hardcoded?

[Sirber]

Well, let's get to the point. The comparison is made on 8 different frames of the anime:Finding Nemo is not anime... :sly:

[pieroxy]

Well, I see a lot of criticism in there, and am willing to accept it. The way I backed up my movie is not the best, that is for sure, but I encoded 7 times the same thing, so I don't see where my test is invalid.

The thing is that I am willing to put effort in doing reviews. What would be great is for people to tell me where I am missing crucial stuff and what YOU - doom9 users - would like me to spend some time on.

As for the default settings, I understand you don't care about them. That's of course not to say nobody cares about them. Just tell me which settings you think I should tweak and try to review. THAT (IMO) would be a constructive way to criticise my poor (in your view) work.

That's if you care for me to do a/some review you might find useful.

Thanks for the feedback.

Edit: As far as 'where are the bits gone', the fact is that I am aware that the more bits should mean the best quality (when encoding the same thing with the same parameters). When I encode a file in 4000kbps I expect it to be of a significantly better quality than one encoded at 2000kbps. That is if the file at 2000kbps shows any defect of course. I don't see what I could have failed to understand here.

Granted the 'quantization factor' is just nonsense and I will edit it tonight. But the fact remains that the 500kbps file is looking poorly on most parts except some. Thus one could think that the 'bit allocator' (sorry if my terminology is not the right one) could do a better job there.

[Teegedeck]

Thanks for taking the not-too-enthusiastic reactions not too badly. :)
We all see the effort that went into making that review so neat.

As for 'where the bits went', they just went into quantizer 1, the destroyer of bits. ;) You see, we got many complaints in the old days about the second pass in XviD producing undersized files because some people don't understand that the use of the standard matrix restricts the maximum possible quality (and bitrate) of an encoding. So a minimum quantizer of 1 has been allowed again, which does nothing but hush up the fact that many encodings would benefit from using either a higher resolution or a high-quality custom quantization matrix or both. Load your .stats-file into the stats-editor in order to find out how big the movie actually needed to be in order to show all quality that the standard matrix is capable to deliver. I bet it is less than your 2nd pass' filesize. And encoding to a bigger file than that doesn't improve quality in any way. Not with the standard matrices.

I know this is all terribly technical mumbo-jumbo, but on this technically-minded forum, users kind of expect reviewers to be aware of technicalities. Your test is very valuable for us though because it shows how XviD just cannot deliver what it's really capable of if only the current default settings are used. I hope that in the future, XviD presets will make it easy to get more quality out of it without knowing how it all actually works.

[raeltheimperialaerosolkid]

Pieroxy, for what I can see, you reached out the same results I had found in my own tests: XviD does its best (at "standard" resolutions) in the range between 900 and 1300 kbps. Less is too low, more is usless.
Coming to your test: Did you used "Adaptive Quantization"?
I usually encode anime and, as Sirber says, Nemo is not anime, but I would say that it isn't "live action" anyway. AQ (in my experience) causes a lot of problems with wide monochrome areas (feature present both in CG and Animation). For compatibility with Stand Alone players I never use Qpel and GMC so that I can't give you any support on this side and you didn't tell if you used them in your encode. I found very useful the presentation of your test the frame-zoom comparison is quite easy looking.

[mod]

test is very valuable for us though because it shows how XviD just cannot deliver what it's really capable of if only the current default settings are used.
I agree, my question is: why those default settings? If I missed something I'm sorry, but after years of encoding with my sweettings your (right) observation made me onlooker.
The default ones are optimized for general purpose encoding, or just something else?

[foxyshadis]

Well, back when my oc'd Athlon XP was new, they would have been perfect; I get close to realtime with a top-end machine on maxed settings, others can get 60 fps with default settings, lesser machines could get 15-20 with defaults. They just haven't changed since then. Nowadays even sempr0ns are significantly better, and dual cores make mine look like a Nerf Starter Edition CPU. You can get 100 fps out of even a single-core FX with maxed settings, or near realtime HD, so why wouldn't you?

pieroxy, I didn't mean to denigrate your effort, it's just that the source needed some sort of cleanup (any filter from gordianknot on low settings would work, even vdub's smart smoother hq would for this). And looking at screenshots is flawed because no one watches video as a slideshow; you have to analyse the motion, blockiness, and detail in a scene as a whole, and pick a screenshot that represents it for viewer consumption.

Both of your tests have highlighted shortcomings of xvid's defaults, which is nothing to snort at. ;)

[dukey]

a better test
might be to encode your chunk of movie (not the entire movie)
5 times using 1 pass target quantizer
first time with quantizer 1 ..
then quantizer 2
all the way to 5

Then compare the results. There is also no 'best' bitrate for videos. The compressability of videos varies massively. Vids with still shots where the people are moving, compress SO much better than non stop action/blowing shit up.

[mod]

@foxyshadis: so it seems that were choosen as default settings the ones which granted the maximum speed. Ok thanks :)

[pieroxy]

Ok, let's settle a few things ;)


The 'where the bits went' was more a rhetorical question than anything else. I know they didn't vanish into thin air. I just wanted to point out that they served little purpose in increasing the quality of the encode.
"I hope that in the future, XviD presets will make it easy to get more quality", my point, exactly. That's why I did this test with default settings. There is (I am sure) numerous people using XviD that don't have time or will to come here and babble about technicalities: They just want an encode done. The default settings ARE important, because that's what some people will use, the least knowledgeable of them that is.
I know that frames comparison is not all. That is why the notes I assign every frame of every encode is based on the frame AND my feeling of viewing the part where it is in the movie. My comments state that in the review.

[sysKin]

Default settings are very good. The only variations, for dvd backups, can be qpel (if standalone compatibility is not an issue) or higher vhq.

I see no other useful settings really, as far as dvd backups goes (not counting credit zones etc).

[Soulhunter]

Btw, it looks like a different chroma-upsampling was used for the encodes...

I guess you've forgotten to set XviDs decoder to output YV12 instead of YUY2!?


Bye

[Teegedeck]

Default settings are very good. The only variations, for dvd backups, can be qpel (if standalone compatibility is not an issue) or higher vhq.

I see no other useful settings really, as far as dvd backups goes (not counting credit zones etc).CQMs...? To my eyes, MPEG and H.263 matrix don't deliver transparency at quant=2 on high-quality sources. Other variantions would include CQMs for lenient compression; for extreme compression; for sports; for anime.
You'll also want to do your first pass at a different quantizer than '2' if you use CQMs with low coefficients in order to reach a steady quantizer distribution in 2nd pass. You'll want the same if you do just the opposite and want extreme compression.
The defaults don't even include VHQ for b-frames...
AQ for strong compression...

The defaults are good. But in order to get out of XviD what it really can do (and makes it a still worthwhile alternative to AVC) you have to delve into the abyss of advanced settings.

[Elic]

sysKin
> I see no other useful settings really, as far as dvd backups goes
Hmm... And how about bitrate curve compression? IMHO XviD is the only compressor that allows to decrease quant on nearly-static scenes and to save bitrate on high-motions/high-details.

[laserfan]

...some people don't understand that the use of the standard matrix restricts the maximum possible quality (and bitrate) of an encoding. So a minimum quantizer of 1 has been allowed again, which does nothing but hush up the fact that many encodings would benefit from using either a higher resolution or a high-quality custom quantization matrix or both. Load your .stats-file into the stats-editor in order to find out how big the movie actually needed to be in order to show all quality that the standard matrix is capable to deliver. I bet it is less than your 2nd pass' filesize. And encoding to a bigger file than that doesn't improve quality in any way. Not with the standard matrices.

I know this is all terribly technical mumbo-jumbo...I'm one of those guys that does not truly understand the mumbo-jumbo, but I really appreciated your post because it gives some insight into how to get RESULTS from the encoder. It seems that the "newbie Xvid" guides, or at least every one that I have been able to find, tell what the settings do, but not the impact of them, or how best to make them work in concert with each other and with the end-goal.

I have found settings that mostly work for me (I make HD XviDs from HDTV caps, and SD XviDs from these and from DVDs) but can't honestly say I know why they work or whether they could be better! I just found out (by using staxrip which pointed this out) that I've apparently in the past made my encodes way bigger than they need to be, both HD and SD.

This business of quants & matrices is a complete mystery to me. :o