Settings for 2 CD rips?

[Latexxx]

I haven't really done much encodings lately, but know I'm in a situation which forces me to do archival quality copy. So, I'm targeting for 2 or 3 CDs with original AC3 audio.
Do you have some recommendations for matrix and other settings? But please, don't suggest 8 consecutive b-frames because need to play this on P3 500 MHz.

ps. I'm going to do anamorphic encoding without resizing (maybe even without cropping).

[Teegedeck]

Well, that's pretty much what I've been doing since I got a DVD-burner. Unresized, cropped (a good idea if you want to decode it on a slow machine), onto half a DVD. That's a bit higher a bitrate than what you aim for; I've also used 5.1 HE AAC instead of AC3 (resulting filesize is about half of the AC3's).

Still I think, you'll get superb results (I dare say, CCEing the stuff onto 1 DVD-R will hardly look better). I've used the same settings as before (highest VHQ, b-frames 2/1.5/1, trellis, gmc, adaptive quantization & quarterpel) - only that I could now use SixOfNine exlusively which makes all the difference (previously I'd mostly used hvs-best-picture). I guess you'd get an average of quant=4.something while I get quant3.something. Have fun, going for 2000kbps is the right decision! :)

Edit: If you get too high quants with SixOfNine, its hvs-version compresses better (but also produces a less crisp result).

[Latexxx]

Thanks for the reply. Just one question, SixOfNine 20 or 24?
Edit: And another one: with or without lumi-masking?

[Teegedeck]

I've used '20'. Also adaptive quantization = lumi-masking.

All that is just what I've used. Maybe SixOfNine '24' gives you a little better compression that you might need, who knows. The perceptible difference cannot be big.

[Didée]

Imagine me, how I am sitting at my desk,
the head buried in the left arm,
whilst hammering with the right's fist into
the sea of tears that is flooding all over the table.


It's

"SixOfNine" (a.k.a. "max=20")

and

"SixOfNine-HVS" (a.k.a. "max=24")


Teege, a big, warm & heartly "thank-you" - for you are the only one who ever managed to call them right. Sometimes, at least ;)


While I'm at it:

Now that ffdshow can correctly decode *all* matrices, it is definetly time to make a change to SixOfNine-HVS:

The DC cell of the Inter matrix was ever meant to be a "10"! The "12" was only a workaround for ffdshow's bug (and not even fully sucessful).

Regarding the manner in which these matrices were spread: how should that change make it through to public? I've little hope ...


- Didée

[Teegedeck]

Oh, my tears of joy mix with tears of shame... :o

Edit: Posting a new thread that announces 'definite and official' versions of your matrices and provides them for download should easily do the trick, I believe.

[Didée]

Sure. What I fear is the "snowball effect" - within short time, we'd have some dozens of threads where people offer their beyond-ultimate set of matrices.

Hmh ... perhaps one sticky where people post links to their matrices, together with the "last updated" date? Could be a nice solution, but requires a lot of discipline, too: the posts should only be edited, then. It would not be a thread for discussion. No-one likes to browse through monster threads (like e.g. the ffdshow thread) just to hunt the tiny little piece of information s/he is searching.


- Didée


P.S. What shame, Teege? You are the good guy :)

[Teegedeck]

Stickies are those things that people tend not to read... :p ...at least they don't tend to check them for updates!

I think you'd have a good start if you contacted LigH who to my knowledge has accumulated the biggest collection of matrices for his matrix-editor. I'm certain he'll like to update his collection which in turn will be downloaded by many people. I think his collection could also be linked to in the FAQ...

[Sharktooth]

Originally posted by Latexxx

Do you have some recommendations for matrix and other settings?
both SixOfNine matrices should be good for 3 cd encodes.
However i suggest you to try also my EQM matrix (you can find it in the Custom matrix comparison (http://forum.doom9.org/showthread.php?s=&threadid=76021) thread).

[Latexxx]

Thanks to my crappy P2 500 MHz, I'm currently encoding using SixOfNine-HVS @ q3. The beginning my source is super-dark and super slo-mo and the average bitrate of the first 11 per cent is just 935 :cool: but it will rise definately later. According to some test clips, which I done, the quality is superb and Mplayer can play it flawlessly on my PC. :D

[Soulhunter]

Originally posted by Didée
Sure. What I fear is the "snowball effect" - within short time, we'd have some dozens of threads where people offer their beyond-ultimate set of matrices.Damn, me feels so guilty now... :D :p :o


Bye

[loni_blues]

@Teegedeck

In your experience, which is the recommended minimum bitrate to use the sixofnine matrices? Would you stick to hvs-best-picture for lower bitrates?

Regards,
loni_blues

[Teegedeck]

Well, it doesn't only depend upon bitrate, it depends on the relation between bitrate and resolution and frames per second. I mostly use it for 25 fps unresized anamorphic encodes at around 2000 kbps. If you use lower resolutions you should get nice results at slightly lower bitrates, too. I recall Didée mentioning he often used it at quant=4(?), and I also think that it still looks good at that. Instead of using SixOfNine at quant=5 you should be better off using SixOfNine-HVS which is more benign in compressibilty - but remember what Didée said about SixOfNine-HVS above:
The DC cell of the Inter matrix was ever meant to be a "10"! The "12" was only a workaround for ffdshow's bug (and not even fully sucessful).

[Soulhunter]

Originally posted by Didée
The DC cell of the Inter matrix was ever meant to be a "10"! The "12" was only a workaround for ffdshow's bug (and not even fully sucessful).Shouldn't it be a "16" instead of a "12" to bypass ffdshow's bug ???


Bye

[malkion]

Shouldn't it be a "16" instead of a "12" to bypass ffdshow's bug ???

Fans are dying to know...

16 is what the mpeg standard matrix uses this much I know, but is 16 really what it has to be? More importantly I suppose, is 16 really needed for stand-alone DVD player support, or can it be anything (such as 8). I guess it's the same question, right?

p.s. And how do you use the Quote function to specify as "Originally posted by..."

Thanks and best wishes...

[Teegedeck]

Oh, ah, just thought of one minor thing that isn't XviD-related but archival-quality-related: GKnot puts an "undot()" into all scripts by default. I would comment that out.

[Fox Mulder]

What's wrong with Undot()?

[Didée]

Originally posted by myself
The DC cell of the Inter matrix was ever meant to be a "10"! The "12" was only a workaround for ffdshow's bug (and not even fully sucessful).
Originally posted by Soulhunter
Shouldn't it be a "16" instead of a "12" to bypass ffdshow's bug ???
... to bypass ffdshow's formerly bug ;)

In fact, yes.

What I figured by that time, assisted by trial & error, was this:

"16" as DC component for the Inter-matrix is (was) safe in any case.
But, this is sub-optimal for hi-Q results, especially when such a matrix gets used in 2-pass scenarios along with higher quantizers. In the given case, the "16" quantizes 60% more than the "10" that should be used. For a frame @ q5, that means the DC component would be quantized by "80" instead of "50". Remember, the DC component quantizes the "average luma" of a given block, and this is the reference for all other de-quantization done for the given block!

In practice, the differences would be hard to spot ... except for flat, uniformly colored surfaces. Just these would be much more prone to blocking-artefacts, if the DC component is quantized too much.

So. Having come to know this, I tried, and tried, and tried ... and finally found that a "12" for the Inter DC component *seemed* to be safe to use with ffdshow, because down to this particular value, I couldn't find any resulting errors in my tests. (But later, I was prooved wrong (http://forum.doom9.org/showthread.php?s=&threadid=72199&highlight=Does+produce+ffdshow)).
That's how I came to use "12" for the DC component in the SixOfNine-HVS matrix set. It seemed safe, and is much closer to the desired value, resulting in better coding of flat areas.


However, now that ffdshow can decode all matrices correctly, I strongly recommend to put that "10" in there.


Originally posted by Teegedeck
I recall Didée mentioning he often used it at quant=4(?)
And still do, yes. In particular, I like it for constant-quant encodes, where q2 / standard-mpeg produces too high bitrate for my taste. SixOfNine(-HVS) at Quant=4 is very similar to Standard-mpeg at Quant=2.5. Which doesn't exist, of course. And I don't like those q2/q3/q2/q3... quant fluctuations - it *is* noticeable, and it unneccessarily wastes bits that a real constant-quant doesn't.


Suffering from a hard 2-days trip with little sleep and much drinks - hope I was understandable, after all ;)


- Didée

[Alxemi]

Posting a new thread that announces 'definite and official' versions of your matrices and provides them for download should easily do the trick, I believe.

Or you could atach them into your signature as the way solhunters do. :)

[Teegedeck]

Originally posted by Fox Mulder
What's wrong with Undot()?
Nothing. It's very nice for 2-CD encodes where it removes very slight noise in the picture and gives a nice boost in compressibility. But if you aim for an encoding that resembles the original close enough to call it `undistinguishable', upon closer inspection you'll find that it removes some little details and makes textures a teensy bit less sharp. Which you don't want and don't need to let happen if you go for 2Mbps, anyway.

[vigi_lante]

OK, about matrix, we are fine, but what about other XVID settings (like b-frames etc) for encodes at around 2000 kbps ? :D

[Teegedeck]

Errr... Haven't you read the beginning of the thread? I mean, before your post it was just one page long...

[superdump]

May I ask why you use AQ? *checks out the AQ thread* I don't think it'll do anything positive for very high bitrate encodes. From what I recall of talking to sysKin when he and MfA were discussing what to do with it it's just lumimasking and increases quants on very light or very dark blocks where the eye shouldn't notice so as to save bits, how is this going to maintain closeness to the original?

[Teegedeck]

I do unresized encodes to 2240MB with the SixOfNine matrix and 5.1 HE AAC. Using AQ makes for the difference between getting mostly base quant=3 (with AQ) and mostly quant=4 (without AQ - a little bit exaggerating perhaps, but you get the point).

edit: BTW, if Didée himself, who designed the matrix, uses AQ I feel all the safer doing the same.

[superdump]

Fair enough, but which looks better?

[Teegedeck]

I had thought this concluded logically from my choice. Base quantizer=4 looks much worse to my eye than base quantizer=3 with some individual blocks at quant=4 and 5 thrown in. That is why I prefer having less frames with base quants=4 but adaptively quantized frames at base quant=3 instead.

The change in quant-distribution is a bit more complex than that of course, but it pretty much boils down to that.