I removed vmax_b_frames=2, scplx_mask=0.24 and naq from my settings.
You may have a different opinion, here are some BV screenshots.

Using lmin=1:scplx_mask=0.24:vrc_eq=tex:naq:vmax_b_frame s=2


Using lmin=1:vrc_eq=tex:vmax_b_frames=2


Using lmin=1:vrc_eq=tex


Using lmin=2.49:scplx_mask=0.24:vrc_eq=tex:naq:vmax_b_fr ames=2


Using lmin=2.49:vrc_eq=tex:vmax_b_frames=2


Using lmin=2.49:vrc_eq=avgTex:vmax_b_frames=2


Using lmin=2.49:vrc_eq=avgTex


Using lmin=2.49:vrc_eq=tex



Bilu

Do I see that right .... even on a Movie encoded at AVG Bitrate 6065 (first one) a average Q of 7.52 results! The solution is to find a dynamic Q which moves between 2 and 4 in average - and its possibe as I saw yesterday that my movie at even ca. 1700kbit average resultet in a average Q of 2.

No contest spirit at all my friend bilu .

Maybe I misunderstood your screens above?

Inc.

PS: As I saw in your script you posted that you use real interlaced sources ... maybe that can be a problem as it would be much more sensitive according to quantisation? But still it shouldn't rise that high in case of 6065kbit average bitrate.


AND ... I do see: The more the settings lower the avg bitrate the less average Q will be the result
Thats totally unlogic, means the more you lower the avg bitrate in regular the more avg quantisation factor you get ... if encoder is set optimal.

bilu,
sorry, I'm in a hurry now. Just wanted to comment a couple of things. I wish I have time to look for a place (Lycos) to upload my graphics, to make easier comparison.
First comment: it looks like you loaded in BitrateViewer each test, without closing the program. I've noticed that in my BV version, I get different results if I load one test after other than if I close and open again the program for each test (yes, I know, it's , but believe, it happened to me several times)
Second comment: your results look a bit surprising, and don't think is that we all desire (I agree with inc comments). I got different results (tht's why it could be better to upload my graphics). Please, try again your settings with your test source, but use vqmin=1:mbqmin=1:lmin=1:vbitrate=9800 instead (don't know the naq effect, if you can try with and without).
And if you're so kind (I know you are) please post your results, if you think it's of a help.
I'll post properly when I have the time.
You're always very helpful, man

Avg quantizer between 2 and 4 wouldn't be DVD compliant is this one
9800 Kbps wouldn't be enough.

Screenshots remade (closing and opening BV before each shot) and adding two more near lmin=1:

lmin=1:vrc_eq=tex:vmax_b_frames=2
lmin=1:vrc_eq=tex

Maybe not using naq makes a difference, we'll see.
About interlaced sources, my PAL "Abyss" source is interlaced, avg bitrate 970 Kbps in the small samples I tested.

I know, looks strange. But remember that b-frames are not as efficient with interlaced as with progressive sources in terms of motion estimation.
About scplx_mask, seems better to avoid it.

If I could transfer this interlaced blended-field high-action anime (45 MB) to you I'd challenge you to try to tame it

If I can dominate the extreme then I can do the easy ones as well

Just posted new screenshots near lmin=1, have a look.


Bilu

If you look at the screenshots using lmin=1 and lmin=2.49 you can see one interesting behaviour:

scplx_mask and b-frames lower bitrate and quantizers when lmin=1.
With lmin=2.49, it works the opposite way.

Why? I think the answer should be something like this:

lmin is the minimum langrange multiplier, a concept like "your ideal quantizer should be around here". It doesn't limit the minimum quantizer and respects the motion estimation and comparison functions, but works like an "influence".

So it influences quantizing enough during the whole stream to make both b-frames and scplx_mask look like they've done a poor job.

Just speculating, of course.


Bilu

@ Bilu
As said even if Q avg of 2 is outputted, my max Bitrate peak was WITHIN the DVD compilant range. And Quantizer of 2 or 4 by this is DVD compilant.
Why do you for approching correct settings use a Interlaced movie?
ok, you will try to work on worsed circumstances to approach best for everything but I think we first should get progressive sources under control.
BTW.... this Abyss PAL one .... is that a capture?? Or do you just refer to DVD2AVIs report window on the right side were almost everytime "inerlaced" is shown .... which is not true in general.

?? Abyss - Anime? Is that a anime remake or ..? Please clearify that

To me it seems you REALLY do your tests on the worsed case which could exist .... a 23.976 to 29.97 telecined and after this back to 25.000 fps fieldblended Source The little "hell" conversion made by a converter box without respecting conversion rules --- TV stations sometimes do use that

That's for your source, not mine. The source I've tested here is an anime sample of Neo Genesis Evangelion, a field-blended high-action interlaced intro which pulls the bitrates to the extremes you can see in my shots.

As I said before, I think that ones are easier. I've made quite a few progressive tests already. But dominate the extreme and you'll rule them all

"The Abyss" PAL version suffers from field shifting in some parts.
Nothing that using Uncomb() or Telecide(post=0) wouldn't solve on a Avisynth source

The screenshots are not from the Abyss but from Neo Genesis Evangelion.
It's an NTSC stream.


Guys, sorry if you thought that this was from the Abyss, I didn't make it clear


Bilu

vqmin=1:mbqmin=1:lmin=1:lmax=4:vrc_eq=tex:vbitrate =9800:vrc_maxrate=9800 is probably what you're looking for.


Bilu

Bilu

1. (just for you personally) Try to get that 25 field blended anime back to its original 23.976 progressive! state using Didées Restore24() or Sharfis_Brain's Unblendpattern() function at doom9.de ... just a hint, by this you will gain a LOT of quality!

2. Now Interlaced Cracks would hit me but ... I saw that even if encoding interlaced (mine was also a fieldblended testsource) as progressive, means idct NOT set etc. ... the stream comes out PERFECTLY at a average Q of 3 ... after this I patched the stream using restream to interlaced specs ... normally thats not the way BUT I recognise what I see and it resultet very nice.

3. I see in your last BV postings that the curve does (IMHO) not output a better Quality compared to CCE as CCE/TmpgEnc do both also output high Q peaks .... so mencoder for me was meant to give me beside fast encoding also a quality gain .... thats why all the pain of testings in here do make it worth for me


Inc.

PS: This night Im going to put a new HD in my PC to install a full Linux system on it , ... I want to see how it behaves on linux .....

.... And thats the point: We should "discover" WHICH windows build does work korrekt! As Vbitrate should be the one to predict with. Im very satisfied with the build I do use ... but on the other hand I cant encode without demuxing to m2v afterwards.

This babe is native 29.97 fps (not a telecine) and the field blend is like this:

A B C
A AC C

where B is an original field and AC is a blend from previous and next fields. IMHO you can't make a B bottom field out of AC, although I had already found ways to deinterlace this bitch in my Avisynth times.

But now I'm fighting for automated encodes and trying to avoid any sort of post-processing related to deinterlacing. Specially because of hybrid sources.

It can happen to be nice if it's just a shifted field since it doesn't hurt motion estimation that much. But try doing that over a telecined source as I did in the beginning and you'll see the ***HORROR***

Look at my last post. Your goal is different from mine.
I'm trying to do an all-proof generic encoding process.
You're looking for the best possible quality.

About speed, my last settings give me 12 fps on my PIII-500 on PAL sources

On NTSC sources, softpulldown slows it to 10 fps. Still fast


Bilu

I got exact an fieldblended conversion out of 29.976 telecined.
Just a fieldshifted I do NEVER encode as interlaced as it can be restored in PAL by Telecide(guide=2,post=false) even if its dynamically phase shifted
Well ... but anyway .... fieldblended streams out of 29.976 animes (which have never been shot on real video "could" be also restored by using unblendpattern() ..... but I don't want to start getting out of topic now

Yes, I know

This anime is native 29.976 fps and is still 29.976 fps, is hasn't been framerate-converted. Of course better quality can be achieved and I DID achieve it with TomsMoComp(-1,-1,0).VerticalReduceBy2() on this stream.

But that requires a stream analisys for every movie and my goal is automated cross-platform reencoding.


Bilu

This thread is obsolete. vrc_minrate affects avg bitrate.

Extensive tests without vrc_minrate
http://www.kvcd.net/forum/viewtopic.php?t=10229