Hello

The correct information on XviD settings comes from the few developers (often when they are contributing to forums) - this information gets rewritten and distributed on the forums and personal guides etc (type VHQ into Google or most forums and you will be told that you MUST use VHQ to increase the quality - not true in my case - if you read the explanation of VHQ given by one of the developers).

Everyone has their own objectives in using XviD (fitting video onto CDs etc). I have my own specification (below) and I would appreciate it if people could comment on whether I am 'getting it right'. I am fully aware of posts asking for best settings - firstly, I believe I use the best settings and am yet to see anyone else listing them, secondly, if we are not arguing about what the best settings are, there are only a few other things to talk about... I would like help from other XviD users who have also spent time researching. Please correct me so I get better - Thanks!

My specification:
I have 480GB of hard disc space and a DVD writer - filesize is not an issue - I am only interested in quality i.e. a perfect(!) conversion of MPEG-2 to MPEG-4.
The XviD should be significantly smaller than the original MPEG-2 (otherwise the MPEG-2 may as well be kept).
I encode everything at 640 pixel resolution (640x480 or 640x352).

When file size isn't an issue, quality is mainly about the quantizer. The ultimate quality setting (using 1-pass with the quantizer set to 2) produces a 4:3 video file that is nearly as large as the MPEG-2 version.

Therefore, I think the following are the next best settings within my specifications - please add comments

A 2-pass encode:

Key frames (I-frames) are encoded with a fixed quantizer of 2 (max = 2, min = 2).
Delta frames (P-frames) are encoded with a fixed quantizer 3 (max 3, min = 3).

Quantization = MPEG - There is a high bitrate so the picture will be sharper than H.263
Quarterpel = On - The bitrate is high enough to warrant using Quarterpel
Motion Search Precision = 6 - The best quality
VHQ = Off - This was the hardest to find out about. Nearly everywhere states it MUST be used as it gives better quality. Not correct - as one of the developers quoted: VHQ doesn't change the quality - it reduces the bitrate at a fixed quantizer Syskin at Doom9
As quality is reduced it is not used.

Chroma motion = On - Acts as an extra motion search precision
Keyframe spacing: 25 - One I-frame per second (PAL)

All other settings (GMC, lumimasking etc are to reduce the bitrate further with a slight quality reduction - they don't meet my specification so none are used).


That's it - thanks for reading. Now please add your opinions!

Skywalkerjen

A couple of comments, first there is no best settings you will never find any settings that work universally well on all movies. Secondly, your information on VHQ is incorrect. It does not reduce bitrate at a fixed quantizer. In the first implementation of VHQ it did but it was found that a lot of times this did reduce quality. So syskin implemented VHQ so that it is rate distortion based so it will always increase quality as well as lower the bitrate. Also if you are encoding from a dvd you may use higher than 640xX you can crop all the black borders and encode anamorphically if you don't care about how big it gets. Generally I like to encode at about 704xX. I crop the black borders and resize it with lancosresize. I find this works well for me and it is a fairly decent compromise. Also quality is not mainly about quantizer; I can show you lots of frames which have been encoded with quant 2 and look like they are about quant 31, shocking? Not really. Also I can further confuse this by using custom matrices, and make quant 2 like quant 1 and so on and so forth. Forcing all the quants to 2 and 3 is very very bad when you use 2 pass if you use two pass you must let it do it's job. If you constrain it like this there is no way it can. You may get constant quants (acutally probably not) but more likely it will be constant crap. XviD will not be able to make any of the right decisions it wants to make. If you are concerned about some frames with really high quant let me assure you this doesn't happen often. Usually when I encode a movie almost all like > 90% of quants are below 5 with none above 8 and this is without restricting XviD's max min quant whatsoever. Your other settings aren't bad although usually mpeg matrix isn't noticable sharper than h.263. Also there is nothing wrong with encoding at a one pass constant quant if you have the space to deal with it . GMC doesn't reduce quality then again it doesn't increase it a whole lot either so it is better to leave it off in most cases since it is slow. And chroma motion searches the chroma plane for a MV I believe.

That is a hell of a contradiction. Its a conversion, so there will be q quality loss, no matter what. It is very easy to make the encoded video exceed the original MPEG2 filesize by far and still it won't be superior to it.

As filesize IS of concern to you, and considering the unavoidable loss of quality, IMO you should take the aproach of sacrificing only as much quality as you are willing to, plus using the bitrate economizing techniques both in the codec settings as well as in image processing.

I'd be encoding the video at full anamorphic resolution, aplying cropping as needed, using the AviSynth method with fluxsmooth to filter out noise (i've seen cases were this halved filesize when using fixed quants). As for the XviD settings I'd be using h.263 (people will tell you to use MPEG or custom but I preffer h.263's cleaner image) and I'd be definitely tinkering with the b-frame settings, as it's THE bitrate saving technique. Yes they will lower the quality as well, but you can reduce this loss lowering the b-frame quantizer (I use ratio 1.0 and offset 1.0, so b-frames get quantizer 3 if I encode with fixed quants 2) and by tweaking BVOP sensivity (for more or less b-frames - normaly they represent more than 50% of the videos frames).

Oh, and I don't see a point in doing 2-pass if you don't have a target size.

You should make some tests to decide what is still "perfect" for you.
Tell us what you used in the end

Sorry, but this is just wrong. You wont gain anything with this, other than faster seeking, but the filesize would be significantly larger.

I would also suggest to use b-frames with ratio and offset both 1.0. The quality degradation is hardly noticable, but it would bring a good filesize gain.

I don't agree with Raptus, mpeg matrix is better for high quality encoding. VHQ mode 1 is also recommended imo.

Thanks for taking the time to make such detailed replies - much appreciated!



No problem - faster seek time is what I am after - can't stand the wait so this I am fine with.




I could have been more precise but I don't think it is a contradiction. MPEG-4 is supposed to be of similar quality to MPEG-2 but at lower bitrates. I am not interested in gettting the file as small as possible - just significantly smaller than a VOB file - you can do this but still have a pretty large file!



I am in effect doing a refined 1-pass encode. 1-pass specifies one quantizer for all frames. I am just refining that by specifying one quantizer for keyframes and one quantizer for delta frames. I have done some tests and can't see this is possible in any of the 1-pass encodes - that is why I do 2-pass without setting a desired filesize.



From numerous tests the codec is trying to encode everything at Q2. To let it do its job would be to not limit the desired filesize. When you do this it encodes everything at Q2. Forcing the quants to 5 and 6 for example would definitely be a bad idea, but forcing them to their maximum is not holding back the codec - only preventing me reaching a desired filesize - which I'm not. And setting them to quant 3 is the only way to meet my "smaller" filesize spec.


Thanks for the info on VHQ- that was exactly what I was after - I will alter the settings to VHQ1 I think. Other info I read about VHQ then was that VHQ1 did increase quality while VHQ2-4 just reduced the bitrate. My Syskin quote was obviously an older one. Thanks for the update (by the way, am I correct about VHQ 2-4?! I read about their use in another forum and that explanation (could) fit with what you said...

Thanks guys

Skywalkerjen

hmm. i have a friend who's considering "future-proofing" digibeta masters (it seems likely that mpeg-4 will be in the next-generation DVD spec) by encoding 9.8mbps mpeg-4 with a standalone box. i would consider this a better way to go about getting the best quality in mpeg-4 then transcoding a DVD. or course, not all of us can get hold of studio released DigiBeta masters

if you have a DVD burner, then there's no problem with burning DVDs is there? if you're worried about the problems of only being able to burn DVD-5 (for now), you could use ReJig or equivalent transcoding software and re-author your discs.

burning DVDs is probably the better way to go... save all that hard disk for your movie projects =)

You can lock quants for I/P/B-frames in all modes. And if You do lock both I- and P-frames (b-frame quant being dependant on I and P quants), 1st pass info has no relevance because you don't let the codec distribute quants (bitrate) as it should, so it doesn't make sense to do a second pass. Agree? With locked quants the final size would be the same for any desired bitrate you enter...

Yes, I definitely recommend using at least VHQ 1.

Tommy, a reason I don't use MPEG matrix is because qpel + MPEG results in the infamous "moving wall" effect, and I don't want to miss qpel.

Raptus, yes exactly if you set that b-frame formula you will get b-frames at the quant you want as well as p-frames and i-frames at the quant you want, then you would use a constant quant 1 pass. This is the best way to do this sort of thing. As for the moving walls thing, it _doesn't_ happen because of mpeg matrix + qpel causes it; it happens in spite of it. If you encode with h.263 you will notice it as well but less so. The difference being that the artifacts which h.263 produces is more blocking whereas mpeg will produce more mosquito noise.
EDIT: BTW from PSNR tests I have seen and from ones I have done VHQ1 does increase PSNR approximately as much as VHQ4 does of course it varies but usually 4 will give a smaller size as well, so in your case 1 would be a good choice since you obviously don't care about filesize. To get the b-frame quant to equal three here is what you do, you use this formula bvop quant = (AVG(past+future quant)*quant ratio + quant offset. So if you want quant 3, 3=2(quant ratio) + 1 ie quant ratio will equal 1.

That doesn't work - you can't individually set the I and P frames separately in a 1-pass encode - I have tested this. Is there some other way?

I quite agree it makes no sense to do a second pass - second pass is to let the codec decide things and I am not. However - don't see any other way of distinguishing between key and delta frames.

Not too clear - sorry. I mean I set separate quant levels as before, but the encode is always a constant quant of whatever the P-frame was set to. This is checked by filesize and also using Classic Media Player with the visualisation on to display quant settings in frames.

You shouldn't really care about the i-frames having a quant higher than the p-frames. In fact you will get higher quality if they are the same trust me. I do understand what you are trying to do. Generally, you don't want too many i-frames either since they are lower quality than p-frames, which is why it isn't a bad idea to crank up the keyframe interval to 250. Hope this helps.
EDIT: BTW there is no way that I can see that you can set the p-frame quant to 3 and b-frame quant to 3. I think it would be not a bad compromise to do a 1-pass constant quality at quant 3 and use a b-frame quant ratio of 1 and use less i-frames say 250, more i-frames is not going to help your seeking enough that it will be really noticeable at all.

Thanks Bonzi

I thought that the key frames were the important ones and had to be higher quality than the other frames as the other frames are built from them.

Setting keyframes to every 250 will give me a seek time of 10 seconds or less shouldn't it?

Is the fact that quality is best if I and P frames have the same quant the reason why you can't set separate values for them in 1-pass?

Your help is much appreciated!

Skywalkerjen

Well you have to have some keyframes but from what I understand the reason why a i-frame is generally lower in quality is because it does not reference anything. Sure future references can be made from it for p-frames and b-frames but you don't reference anything before. Which is why you want them on scenechanges. Your seek time should be very small ie less than 1 sec even with keyframes at 250. Yes the max time between the keyframes is 10 sec but your computer does not take this long to jump to the next i-frame it is not real time that is the whole point of seeking . BTW it is likely that you will get a i-frame less than the maximum i-frame ie scenechange or whatever. The reason why you can't scale the i-frames and the p-frames quants seperately is probably because it isn't that useful really. I mean you don't have that many i-frames in relation to p-frames anyway so why would it matter if you can change the quant for a few frames anyway? I suppose you could get a little better quality if you i-frames were a little lower quant when you have to use them but really it doesn't matter much because of there aren't that many. Another reason why nobody has seperated them for 1 pass quantizer is because you can get away with a little higher b-frame quant on some frames than others; this may sound a little silly but really if you are doing 1 pass quantizer you want exactly that.

As the mods here told you already, thats not a wise idea if you are after perfect quality. Make anamoprphic encodings and use cropping only, no resizing filter. Some light additional noise filtering can be used if the source DVD is pretty noisy ....

Pardon me, but this seems a little nutty. AFAIK, an I-frame defines the beginning of the GOP, and all other frames within the GOP referrence the previous I-frame. With a setting of 250 on a fairly static scene you could actually get 250 frames before the next I. That would be sucky seeking if you wanted to chop a clip out for some reason. Using more I-frames is only a benefit to quality the way I see it.

Perhaps I'm wrong.

Using more I frames is not a benefit to quality. I frames and significantly larger then P frames and even larger still when you them compare to B frames. To many I frames will decrease the quality of the video, or drastically increase the filesize, depending on the codec and settings. If it's a 2 pass encode, it will probably drastically reduce quality. 250 is a good setting.

A thread entitled like this one can lead to no good (except if it was meant ironically) and so it seems. If you want to use XviD in a way as to maximize the quality (given the limitation of filesize) all you have to do is read everything sysKin has ever posted over at forum.doom9.org. Because he actually knows how a codec works. That is all. Not pure accident and not even testing of all the possible settings are a substitute for technical insight.
As a little hint: You gain much more from a good high-bitrate matrix than from limiting your quantizers. Or from using less B-frames, fiddling with the B-frame quant ratio or even setting ludicrously many I-frames (quoting sysKin: "You don't want I-frames.").

Really, forget all that, load the defaults and then go through sysKin's post.

I don't think so - it is an entirely legitimate question. I notice from trawling through these forums that very few questions are actually answered. The only people who "really know" are syskin, Koepi etc, who develop the codec but are too busy to provide the documentation. This I have little problem with - it is free and they are very active in the forums.

As an example, I have read most of the leading forums on VHQ. I am none the wiser on what it is for, nor do most of the people who comment on it - usually contradictory. I am well aware that there is no such thing as perfect quality, but under the specification I have provided there is - the codec has a finite number of settings.



By limiting the quantizers to maximum they have the highest possible bitrate to work with...



I don't use B-frames as I believe they are used for extra compression - not relevant in my spec.



You do if you want seek points - you only don't if you want as much compression as possible - again I don't want that - I already explained.



I have been through most of sysKin's posts on Doom9 already. They are mainly concerned with maximum compressiblilty - the point of the codec. They certainly don't terminate the threads, leading me to believe that the answers are often individualised and don't answer everyone - hence my question.

I am fairly certain I am correct with my setting though as the help provided here has not convinced me otherwise - so thanks very much for all your time guys - especially Bonzi and Raptus for the repeated follow ups.

Best regards

Skywalkerjen

Okay, Skywalkerjen, let's clear up some misunderstandings.

First of all, I cannot believe you that you could have read all that the 'wise' people say on doom9.org and haven't learned from that. Simply because I'm in the same position as you: I am no developer with deep technical understanding, nor do I have enough time at hand these days to read everything everyone posts. But sysKin (if we restrict it to him) has said very clearly what to make of VHQ, and if I could understand that, you can, too. (Put shortly, VHQ always enhance compressiblity and quality!) Maybe you were confused because this was not so with an early version of VHQ. The same goes for GMC (which only really helps when used in combination with VHQ). The higher the VHQ the better. Use as much as you can endure.

Second, and more importantly. You, as seemingly a lot of people, are suffering from a ambiguity in your understanding of video-compression. You have understood that uncompressed is better than compressed and that something looks all the better the less it is compressed. Good. But you've also heard that a codec is all the better if it compresses better. Now what?

Your first notion is true because it is simply a theoretical truth. In so far as a lossless compression must by its very concept always give you higher quality than a lossy codec (like XviD). BUT all this is fine, but you don't want to convert a 6-GB compressed (=lossy) DVD into a 12 GB (=lossless) encode because there's no practical gain from that. The misconception now starts when you find yourself for practical reasons forced to choose a lossy codec but because of your 'the bigger the better'-notion are afraid to use features of this codec which claim to give 'higher compression' (=smaller and worse?). This is wrong. Smaller is not worse if it exactly fits the size you can spend for that file (2 CD-Rs?). You don't want anything that is bigger than these two CD-Rs!

Now how do you attain that goal. Following your concept of how video-encoding should be done, you'd disable all features of the codec which you find distrustful because they are known to deliver 'higher compression' (=worse quality????). Then you look how big the file grows, find it too big and downsize your output-resolution accordingly till the thing fits onto two CD-Rs without looking too ugly. That's the way we did encodes years ago with MS-MPEG4v.2 where you couldn't really influence bitrate at all... Excuse my a bit sarcastic way of describing that, but I do that in order to make argumentative faults hopefully clearer.

The better way to do this, would have been to enable all the features which yield better compression (= smaller filesize without a noticeable degradation of picture quality!) and pump-up your resolution to the maximum till you reach the best compromise in resolution-quantizer ratio. If you still have some air (quantizers at full resolution at 3 or lower) use a high-bitrate matrix like Didée's 'six of nine' matrix and get a real boost in quality. Be aware, 'six of nine' at quant 3 looks much better than the standard MPEG-matrix at quant=2 (which always looks worse than the 'psychologically enhanced' HVS-best matrix, by the way)

I.e., if you want to do what you said in your first post, you don't resize at all but crop to a mod-16 resolution and encode anamorphically with HE AAC sound and put it into an MKV-container.

Gain a visible visual quality improvement at the cost of a not noticeable higher compression. If downsizing resolution was the key to effective video compression, there wouldn't be a need for further codec development.

Really, if you deactivate any feature of MPEG-4 Advanced Simple Profile (b-frames, quarterpel...) in XviD, you could as well encode with MPEG-2. There's absolutely no excuse for doing that, it only made sense for as long as some of these features didn't actually work properly. Now they do. Enjoy all of them.

...and all this learned just from reading posts from the 'right kind of people' at forum.doom9.org. Cool, ey?

bonzi;

no,you most certainly can't!
quant 2 look exaclty like quant2;excellent;31 always look like shit....
but ok,we're talking xvid after all;anything can happen..( )
but i think you misread the stats,graph or something if you claimed this;i think you can't prove this!
i never saw quant2 lookin' bad,and i seen a lot....2 is just doing (almost) the least compression of particular frame,so your claim can't be correct....
sorry mate...
grab drfanalyzer and tell me if i was right....
http://www.geocities.com/analyzerDRF/

@skywalker;don't worry about tg;he's xvid mod over at doom9,after all...( )
one more thing;koepi is not codec dev;he just compiles the codec...
(and if you asked this at doom9 he would flame you to death!)


but let's see if we can be of any help..(rather than discussing stuff that you don't care about;all these folks do 2-31 and you don't care about it;i don't too...)

what is your objection to the quality of the video?

what resizer are you using(and why are u resizing at all?)

1second KF spacing?be aware that many of the mpeg4 advances over mpeg2 are in the way it does motionestimation and how much KF's are spaced(ie. it can space KF more because fo more precise m.estimation)
you may leave it at that,but tdon't be surprised at filesizes or quality;
usually mpeg4 does 10sec spacing or scene change(whichever comes first)...

i-frame quant 2-2?nope,i wouldn't do that either;it seems to be that newer mpeg4 codecs(xvid too) are tuned so that i-frame quants selected stays tuned with surrounding p-frames(btw. i support your decision to avoid b-frames...)
ie codec won't compress all p-frames with qunt2 and then use 31 for ever i-frame...

VHQ?actually on my analog stuff i didn't saw it do anything....

and after all,mpeg4 is only perhaps 10-30% better than mpeg2,and you surely "tweaked" those settings to get them as close as possible( )

if you're interested in constantquant comparisons(so called constantquality encodings),see this
http://www.aussievideosearch.com/cmpeg4.htm
off course,it may not work for you...he's doing quant4 after all...(and getting pretty big filesizes on hires... )

but again this is doing it the blind way;what is the issue you're not ok with on xvid?
best settings i never use would be p-frame 2-2....

anyway,tell us more,or no help for you( )
as far as i'm concerned you may be doing crappy bilinear resizing and coming here asking questions about poor codec perormance on 3-3..and that doesn't speak well about you....not kidding...

also,if one codec is not good enough,there are other codecs,but first you should be able to say what's wrong with this one;you know,there's something AWFULLY wrong in asking "is this good,is this best" as if you don't actually watch your video....
(everyone knows i'm no xvid fan,but when you accuse it about something,please explain it....usually folks say xvid works well on clean (dvd) sources......)


and as a supplement(completely unrelated to this thread) why is h263 bigger in filesize than mpeg matrix?
http://neuron2.net/ipw-web/bulletin/bb/viewtopic.php?t=372

edit;must comment this from tg;

[personal attacks removed][/personal attacks removed]

[personal attacks removed][/personal attacks removed]

Thanks for taking so much time to reply t.g. deck.

The VHQ thing - yeah I read and understood that when Bonzi corrected me:
VHQ1 makes the difference, VHQ2-4 reduces the bitrate.





Yes - but that is not a contradiction. The point of MPEG4 means it matches MPEG2 but at lower rates. If using such an effective compressor, without pushing it to its limit ought to give excellent results.



This is my point and is why I provided the specifications.



The features I am not using (VHQ4 etc) are labelled as giving higher compression with a reduction in quality. In my case (using quant 2) there are no features which can increase the quality and yet provide greater compression. It is fair enough arguing if I may notice the difference, but I am encoding for long term usage - I don't know what I may need the XviDs for in the future.



I was going to keep the AC3 sound - no changes. If AAC has plugins to improve the sound quality, well I have a recording studio that will do the job rather better. I keep the original soundtrack.



That makes a great deal of sense - are you saying I may as well recompress in MPEG 2 using DVD Shrink for example? I tend to agree. That is one of the other reasons for putting the post up. I suppose the main reason for re-encoding is that I absolutely hate the commercial video DVD format. The menus really wind me up. I also haven't found the MPEG2 player that matches the speed of the MPEG4 player. MPEG4 seems to me to be less likely to crash during searching. The format seems to be the way to go and as my storage capacity is only getting bigger I am happy to have MPEG4 video almost the same size as MPEG2.

I suppose the statement - I may as well keep the VOBs is a little misleading - I keep the VOBs but they are still on the DVD. I am after a better format for using the PC as a large movie jukebox.

Thanks for spending so much time though t.g.deck. Much appreciated as I have a fair amount of DVDs ready to go and I am getting the info I require

Cheers

Skywalkerjen

Thanks, interesting read. I happen to be interested in full size ~q4 captures as well. (The test is a little skewed though, because of huge bitrate differences.)


Wow, I didn't know that. I always thought that h263 was trading lower bitrate for sharpness. Nice to see a Powerdesk user.

Thanks for your reply i4004

I am unsure which resizer to use yet - I was told Lanczos, but reading elsewhere in other apps, lanczos is not too good? I am after a sharp one though as the XviDs will be watched on a TV which will smooth the image anyway.



Yeah - I tested that with the one-pass modes and the I-frames do just match whatever settings the P-frames were put at.
But this 2-pass method does seem to separate the I-frames quant from the P-frame.




I suppose I am asking this because I plan on using the XviDs in many years to come - the time it takes to encode them all they had better last that long. I am checking if there are theoretical improvements I can't see on my current set up but which will be evident in 20 years time when my home entertainment system is converting the XviDs to 3D...

Thanks for the links too

Skywalkerjen

As you don't seem to care about filesize-limitations at all, the old 're-encoding is evil' phrase applies and you should keep the DVD as it is and split it onto two DVD-Rs. Any re-encoding means another loss and without filesize restrictions you lack any rationale for re-encoding

Edit: And a 720x..., B-frame (2/1/1.5), GMC, VHQ=4 (what you call 'pushing to its limits'), high-bitrate matrix encode with quants 3-5 will look much better than a 640x no-B-frame, VHQ=1 encode at constant quant=2...

I can prove this, it would not take very long to find some frames that look bad even at quant 2 and this is not an XviD problem you probably could do this with any codec. But you must realize it is a little exaggeration on my part too I didn't actually mean that it looked exactly like quant 31 you can't even really define what quant 31 actually looks like, I just was saying every once and a while you can find a frame with quant 2 that looks poor. t.g.deck is right though if you have access to a DVD burner just split it onto 2 DVDs or try and make it fit onto one without reencoding anything. There isn't any point in making mpeg4 files which are the same size as the original mpeg2.

Thanks for the reply t.g. deck



I do care about filesize (up to 3/4 or so) but the main reason is to get it out of that format (menus etc) and be able to hold everything on a hard disc at as close to DVD quality as possible.



And that is exactly what I am after. I shall now research those settings.

Cheers

Skywalkerjen

[quote]To understand this you have know that we (moderators at doom9.org) threw i4004 off the forum for his misleading and partially insulting posts. [/quote]
to understand what exactly?
that you think that advanced mpeg4 option make a real difference between mpeg4 and mpeg2?
to say that
[quote] Be aware, 'six of nine' at quant 3 looks much better than the standard MPEG-matrix at quant=2 (which always looks worse than the 'psychologically enhanced' HVS-best matrix, by the way) [/quote]
i hope verywetpaint's post helped you there too...huh...

to end this ,i'm quite glad i'm not fearing any mod's boots over at doom9...there..it's always a fear "will mods like this...or..?" on that forum...


[you didn't needed to mention this,but now you did,here we go again;]
misleading and insulting?
1-i said GKnot si a mess for keeping the AR
(few days after that doom9 says that "aspect" tool can be used to aviod slight AR skews when,for example,cropping is done prior to resizing etc...later i learned that GK has a mode where this skews need not appear and apologized to wef...i'm still doing resizing my way,and still gettitng that AR offset=0 every time...)
2-i said xvid has ME bug on motion+noise(floating walls)..by now everybody knows about this....

i will be more than happy to repeat this every time you say how i was kicked out for misleading and insulting posts....
insulting to doom9 and insulting to xvid team,yes they are!
especially when i asked him about "aspect" tool like this;
[quote]ohh,and btw. d9,how's "aspect 1.1" working?
(lol!)[/quote]
here
http://forum.doom9.org/showthread.php?thre...ct&pagenumber=2

few posts after that you said;
[quote]I'm sad that everytime you try to use it, XviD plagues you with exotic bugs that no other user has ever experienced[/quote]

[personal attacks removed][/personal attacks removed]

but i like continuum's reactions to my stuff...
yeap,mate,that hand with yellow cards always looking at me

[quote]
(The test is a little skewed though, because of huge bitrate differences.)[/quote]
i told him he's wasting bitrate;but he wouldn't hear about it( )
but he's not capping at q4;he's encoding at it;he's capping in picvideo....

[quote]Wow, I didn't know that. I always thought that h263 was trading lower bitrate for sharpness. [/quote]
well,it's trading sharpness ok....hehe

[quote]I am unsure which resizer to use yet - I was told Lanczos, but reading elsewhere in other apps, lanczos is not too good? I am after a sharp one though as the XviDs will be watched on a TV which will smooth the image anyway.[/quote]
lanczos=sharpest...no doubts there....good decision...
you're not all that rookie after all...

[quote]Yeah - I tested that with the one-pass modes and the I-frames do just match whatever settings the P-frames were put at.[/quote]


[quote]But this 2-pass method does seem to separate the I-frames quant from the P-frame.[/quote]
i just hate to repeat what everybody else said;but they are correct;what 2pass with fixed quants on such high bitrates?

[quote] I am checking if there are theoretical improvements I can't see on my current set up but which will be evident in 20 years time when my home entertainment system is converting the XviDs to 3D...[/quote]
what will convertt it to 3d exactly?

if you'll be getting bigger screens you wanna avoid lowering the res(in my mind dvd's already look to blurred;the premastered ones anyway...)
also there's a good reason why they are now considering hd-dvd(with wmv9 and h264 as codecs...) specs.....

i suggest this;don't resize,crop black borders(respect the 16 rule)use your codec of choice at typically high mpeg4 rates (cca. 1.5-2Mbit/s average..or if you're dissatisfied you can even go higher)
if you watch this stuff with proper decoder...(for example ffdshow can sharpen your image etc.) it will look pretty good.....
it WON'T look exactly like original(no re-encoding can,and this is mpeg2->mpeg4)
but it can look pretty acceptable.......i'm sharpness freak and i'm using mpeg4 for analog caps;this means it can be done!
i hate mpeg2 because it wastes bitrate;it seems industry also understood this(at last) so they are going for better codec for hd-dvd....
but those thoughts don't need to stop you in waiting for 2layer dvd burners and making exact copies if you find mpeg4 "too-grainy"..(some did....i'm OK with my selected mpeg4 codecs...)


good luck anyway

/ivo

edit;
bonzi;
[quote]I just was saying every once and a while you can find a frame with quant 2 that looks poor.[/quote]
and i said that these are either not quant2's or you saw wrong....
again;i'm VERY interested in seeing such short clips that has quant2 lookin poor...(with xvid or any other codec!)
i never saw 2 lookin bad....that's all i said....

edit2:ok continuum is not a mod here...but he is over at ars...
d9 forum can learn from ars too....

Somehow I don't see how discussing how or why i4004 you were kicked off doom9 is at all relevant to this discussion. Please if we cannot conduct this discussion civily then it would be better that we do not. BTW I can encode a entire movie today at quant 2 just to prove my point but really there will probably only be very few frames which don't look all that great. Really the only reason why I brought it up was because I was trying to illustrate my point that average quantizer will _not_ dictate what kind of quality you get. I suppose it is a rather poor example if you want to take it to the extreme.

sure it's not;but if tg says i was "misleading and insulting" i feel the need to explain what was "misleading and insulting".....i thought i said it already,but apparently some didn't got it...


you will notice that i actually never start these fights with tg.(i said it's laughable that b-frames and gmc (and what not),brought mpeg4 over mpeg2...that's simply not truth...mpeg4 was already better....if that insulted him,i'll be insulting all the time...)...but i do try to put things where they belong...

it's tg's folklore to attack me,and you know i'm not just staying here watching it happen... but let's forget it....(if he asks for it again,he'll get it,again...word by word...)....let's see what will skywalker say;will he like 720 better than 640....etc.


sure average quant. won't...but fixed will( )
skywalker used fixed 3....


good mods admit their mistakes....

tg

and 720 with NO b-frames,NO gmc,NO VHQ will look even better....
(and some time will be saved on encoding)
the last part of the sentence is also suspicious;
quant3-5 on 720 versus quant2 on 640...huuhh.....
i would claim this only after i tried it on few movies;not before that i wouldn't...
quant5 is simply blockier,and resolution difference is not so big so that resolution increase itself will "cover" the blocking.....
if i'm wrong,please let me see the evidence....
b-frames CANNOT bring the quality....they can only decrease it(but they decrease bitrate too..);that's how they work;they are ment to serve as one further step towards better compressabilty=some(hopefully very small) loss of image quality....

i myself have very poor experiences with b-frames on both ffvfw and xvid...(on higher bitrates too)

Ok, I meant that there is no need for _anybody_ to be discussing why or how you got kicked out of doom9 this means you or t.g.deck. It is fine if t.g.deck says he finds your opinion misleading etc. but there isn't any need for anybody to get into the history. Don't worry anyone who gets out of line will be dealt with. But if you retaliate likely you will get it the worst I don't care who started it . I would appreciate if the personal attacks could be stopped thats all.
EDIT: personal attacks have been removed from this thread, one more and there will be consequences, you have been warned. Hopefully you guys have said what you want to say to each other and we can just get back to video, ok?

Isn't this an unfair comparison? The bitrate of q2@640 will be quite a bit higher than that of q(4 or 5)@720. Generally, I try to use the original (=TV) resolution and increase the quantizer for better detail in low motion scenes. (Small text usually suffers from resizing, and besides I'm an interlaced freak... )

(continuum /= Continuum)

I'd like to encourage everyone to research the subject of custom quantization matrices. It is quite fun as you'll find some 'simple truths' are not that universal anymore. You can write a matrix which looks better at quant=5 than standard looks at quant=2 (Didée did that once). [Logically, this quant 5 will have a bitrate comparable to standard quantizer=2...] This one would be apt if you want to re-encode a DVD to a full DVD-R... Didée's '6 of 9' on the other hand looks better at quant=3 than standard matrix at quant=2. (It is NOT compatible with ffdshow.)

The bottom-line is, once you start using custom quantization matrices, you cannot simply belief that quant=2 is better than quant=3. All the more when you use anamorphic resolutions. Most of us discarded average quantizer as an indicator of quality to some extent when we started using B-frames already, of course.

When you compare the effects of different quantization matrices, you'll always have to encode the full movie with both options and compare the results to find out what looks better. I do that on a regular basis because investing an insane amount of time and effort in order to reach the best possible quality is what the fun in encoding is all about, isn't it?

as vwpaint said;


( http://neuron2.net/ipw-web/bulletin/bb/viewtopic.php?t=372 )

and as you said;


as i sidenote i didn't saw any of teh custom matrices producing sharper images than MPEG....do you have any images to show that custom matrices are worth the effort?i don't....
none of the matrices included in rc2 beats MPEG quant...none will spare bitrate...so again;why bother (i'm bothering as for particular bitrates(lower) MPEG rings too much and h263 blurs too much;but this doens't have anything to do with skywalkers sky-high bitrates..(one notion though;sky,what if your hdd (with all the film) dies suddenly..hehe) )

skywalker is surprisingly quiet though....

Sorry I've been quiet - for one you guys are starting to go over my head again and it takes some thinking and researching to work out what you are talking about.



I have that worry all the time! Although they are all backed up to DVD-R I am wondering how 'safe' that is going to be as well. Not to mention how long it takes to copy 1TB of data onto hard discs... I'm not complaining but this is why I want to make the best decision and you guys are giving me loads of info for that - thanks!

I will try and post something more constructive tomorrow evening though, although I am thinking that it looks like I will be "transcoding" (if I use the word correctly) the MPEG-2 over to MPEG-4 while only saving about 10-20% file size.

Oh:

Are you meaning that the increased resolution with the slight loss in quality from using the advanced MPEG-4 features is better than a lower res encode without the advanced features? This I had not considered

Skywalkerjen

Yes, higher resolution (or more details) and more advanced compression techniques is what I meant.

It is just the thing we'll see with HD-DVD as the successor of the DVD, soon: HDTV-resolution, encoded with more advanced codecs like H.264.

The idea is to transform smaller filesize into better visual quality: For example a video encoded with B-frames will be drastically smaller than the same video without B-frames. But it won't look better. But if you encode the video with B-frames at a higher resolution and/or with a high-bitrate quantiation-matrix you'll have a video that actually looks better than the video without B-frames at the same filesize.

But if you want to save only 10-20% per movie I strongly advise you to use a compressed domain transcoder like Nero Recode instead of XviD. You will get a result that is indistinguishable from the original in just 10-30 minutes.

Edit:
So which options are a must?
B-frames, 2 of them, offset=1, ratio=1.5 (defaults)
Adaptive Quantization (very safe and effective now!)
Quarterpel (this one actually enhances visual quality)
Trellis
VHQ=4 (if you are patient enough)
GMC (the least important on this list, use only in combination with VHQ)
for high-bitrate encodings a matrix like Didée's 'SixOfNine' or the HVS-version of it that he just posted on doom9.org

And after all this discussion we're back to (basically) exactly what the XviD 1.0 Guide at doom9 already says. t.g.deck is correct though, as these are settings you will want to be using.

this thread overall has a HUGE problem;
i'm constantly getting a picture where skywalker is only encoding his stuff,without ever watching it.....

i asked(i guess we all did in this way or another) "what's wrong with your xvid clips,skywalker",but he never replied....
if we don't know what's wrong,how can we shoot for beter quality(better than what???)

perhaps he should do this(encoding the same clip (or avs script) in all cases..presumably 720xXXX clip);

encode1-normal 2pass xvid 2cdr(or 3dcr) filesize target(wide quants span,but high target bitrate)
encode2-1pass C.quantizer 2(or 3 or 4..or all of them)
encode3-1pass quant.span 2-8(2cdr (or 3cdr) target bitrate)

if skywalker can't differentiate any of these,then he started an useless thread....sorry skywalker,it must be said like it is!

well known fact is that the higher the bitrate the less you need 2pass anyway....and 2-3cdr targets...huh...that's a LOT for mpeg4!

i think he should do such "blind" test......and pick the method which has highest quality/shortest encoding time......if a filesiz eis a parameter then it's a paramtere..if it's not then it's nott (but as i said that hdd may go bad.....it's not a nice feeling,let me tell you...good thing you do dvd backups,skywalker)

this is where i'm signing off....i can't help the man which can't say what's his problem....(as i said in previous posts)

Nothing is wrong with my clips.
I asked a straight forward(!) question at the very start:
I gave you my specifications - which were pretty simple, and asked if my settings gave the best quality for those specs.

t.g.deck has given me some very useful info - specifically going along the increased resolution path.

As I am not planning on handling and converting so much data on even a three year basis, I am wanting these XviDs to look good and be useable for at least ten years or so. As storage space, bandwidth etc is getting bigger I am happy to encode XviDs that I can't all store at once right now.

I basically want a future proof format for my video as it takes so long to encode them. Due to the filesize issue, I went along the path of asking which settings would give me the most data while still being a viable alternative to MPEG-2.

My main questions throughout the thread was about VHQ.

I am pretty sure the consensus is to that VHQ4 doesn't reduce the quality in my case.

Could you be more specific about why I shouldn't put the keyframes at the higher quant than the delta frames though. Because if not I will do a 1-pass encode and get started on them.

As for my visual interpretations of my encodes - I can't even tell the difference between constant 5 and constant 2. I just want to future proof them because it is such an outlay of time (and money on DVD-Rs).

The thread is not useless though as I asked a legitimate question which I much appreciate all the time everyone has spent in helping me.

It boils down to this: Use the settings that are recommended in Doom9's guide. If you are an avid reader of the forum, adapt those settings when people like sysKin, Koepi or iago say they've found improved settings. Only when you think that you understand how the codec works and can clearly distinguish differences in visual quality (good eyes and excellent monitor are a precondition) you are maybe ready to experiment yourself. But this should not be necessary at all! There are no mystical 'tweaks' to be discovered. This is not Nandub or anything. The XviD-thing is pretty straightforward.

You could make a case whether or not you want 2 b-frames it comes down to personal preference really. I would _not_ suggest adaptive quantization though, it will not increase quality in this kind of encode actually there aren't too many situations were it should be recommended. Basically if you are going for low bitrate encodes and you need to shave off some more bitrate at the cost of some quality where it shouldn't be too noticable then this is for you. But for sure don't use it on a constant quant encode. Qpel should be good, if you don't mind that it will take a little longer to encode and take more cpu power to decode, it will increase quality on most material. VHQ4 can be used as well like you said if you are patient enough. GMC probably won't be that helpful and will slow down your encode a lot, I wouldn't enable it.

BTW, I don't agree that AQ is not apt for high quality encodings. Though this is close to personal taste and belief and not to be confused with solid fact!

Have a try and watch the quantizers in a video done with AQ with ffdshow's --> visualizations --> quantizers. You'll be able to tell that there now are many, many frames where AQ doesn't change the base-quantizer at all, and where it does it does so mildly. And I've never seen adverse effects in parts of the picture that were more highly quantized, so I guess the AQ equation works fine. And still it gives a huge saving in bitrate. Personally, I don't think a video with lots of frames with an average quantizers of 2,48 (for example) looks worse than one with frames more often changing between 2-3 (again an example). More likely the opposite. Perhaps one should check whether not using AQ makes for bigger floats in quantizers during 2nd pass. Maybe I will.

Given that almost all encodes are aiming for a specific filesize I would always recommend AQ.

Edited to differentiate more clearly between fact and assumption, and a blatant error edited out. Sorry, shouldn't write that late

Edit: I'd like to add that using AQ doesn't do wonders in reducing quantizer fluctuation as you could read in the '1pass vs. 2pass' thread. But it 'reduces' bitrate in the eyes of the 2-pass-algo (as anything bit-saving does) and thus base-quantizers become steadier.

Also I collected some more thoughts on quality. The facts about XviD's features stem from Radek's posts, those about matrices from Didées. But naturally the order of steps used for reaching the best quality only represents what I perceive as quality:

It is wise to start out with all fancy extras in XviD activated (quarterpel, b-frames, vhq=4, GMC, trellis...). Some of these extras of course do less for their processor-time than others (GMC, vhq>1), but quality-buffs will want them anyway. Especially vhq which doesn't reduce filesize much above vhq=1 but increases visual quality. Use a high-quality matrix that produces a reasonable bitrate (hvs best-picture).

Then, if your encode would produce a 2nd pass of quant=2 or even saturate, set a higher resolution until you reach full resolution (anamorphically).

If your 2nd pass still has quants from 2-3, go for an even more hq-matrix (ATM I'm very fond of Didées' SixOfNine matrix and also of its hvs-version (a little less tho).). With these matrices (and the one mentioned in the next paragraph), the first pass should be done at quant=3 (in the 'zones'-setup - don't forget to change this back to weight=1 for the 2nd pass...). This will disable 'fast first pass', of course. So enable 'turbo' for the first pass only instead.

If you still get quant=3, think about using Didée's half-insane (semi-insane) matrix. This will get you damn close to the original quality (at about half its size, I guess). If you've gotten to that option, I suppose you use a very clean source.

Still not content? OK, then, and only then, it is time to drop advanced features. You could try skipping aq, trellis and b-frames. Not necessarily but probably in this order - I really haven't had the opportunity, yet, to test this out....

Good point. There are dual-layer burners now for a premium, but most films will fit in 4.5gb if you remove all the crap like extra soundtracks, directors chats and 'special features' with a decent ripper. When I buy new DVDs, they end up as a single huge vob on my pc, which I can keep or reencode.