Phil,

"Cuts off too much ..." means if the values get higher and higher reaching the right bottom corner the more high frequencies will be cut, means in easy words, the picture will look unsharper.

Do your test on DCT filter as I described here.
http://www.kvcd.net/forum/viewtopic.php?t=8551

The dct filter does almost the same, he perfoms a DCT (as our matrix also does during encoding) and does afterwards a InverseDCT (and that will be done when our KVCDs will be played back by the mpeg decoder as the matrix also will be present in the encoded mpeg stream! you simply can use "restream" to suck out the matrix of a m2v or VOB)
In easy words (cracks would hit me now) its like a chiffring and dechiffring (I dont no other word) where the chiffred code is very eay to encode seen in compression. And when while playback the dechiffring performs, it could cause rounding errors.

Do a test and use Restream, ... open a m2v encoded with notch and allocade manually to the stream the std. matrix in restream .... when playing back, trash would come out ;-)
You know what I mean? Its like a Key to get the right INVERSE-calculation afterwards.

So why we get blocks, ok, of less bitrate, but we do use Notch???
Ok, a cutting matrix does give the encoder very precise quantisation orders "where" to cut and thats why very much developers of matrixes do say "XXX Matrix, best for AVG about 1200kbit", because if the bitrate goes even below on high bitrate needed scenes the encoder dis-respects the matrix and does his own quantisation party to archive the needed AVG Bitrate or final filesize and this in a very very agressive way where YOU don't have the control over that quantisation anymore.

And our little friend ffvfw mpeg2 seems to like playing its own quantisation party very very fast.

The last 5 Days I also did tests using diff. Matrixes where for example only the high freq. will be cut straight and Im still in.

If you remember I saw in Kwags samples that the result is sharper!
Means the high frequencies won't be cut that much (8bit values! also at max cut), ok .... therefore the plainsurfaces and mid-detailed parts in the image could be suffer more ... and in case of ffvfw the plain surfaces do suffer more. Its just a thing what the encoder prefers to do.

Every encoder gots his own way of quantisation process and thats why CCE TmpgEnc MCE ffvfw do output diff streams.

But every encoder does come with his own pro and cons. So everybody has to choose what he wants on pros and what he can tolerate on cons.

;-)

Just for freaks if you want to try diverse Matrixes collected by Morpheus of doom9.de
Heres The Thread
http://forum.gleitz.info/showthread.php?t=82

(I just quoted the DVD relevant ones)
"Just" copy and paste :oops: :lol:

PS: Tonight I will do tests on ffvfw using "Angel 1CD 2.35 avg800 max1500" as it only does high cutting on high frequencies, ... we'll see

Hi Inc,

Remember that almost every one of those matrixes (high frequencies), will be truncated by ffvfw :!:
The MAX value for ffvfw is 63, so most of those matrixes will be useless with ffvfw.

-kwag

You mean "with that encoder" ?
Because there is not reason in the mpeg codec protocol that you can't do that. And all encoder can actually do CBR.

Okay, that's the same that we have with too hight value in DCT Filter. Thanks.

Acutally I didn't wait for this post to know what how DCT filters works :-)

scramble/unscramble :-)

Just make me think about something : WHO actually set the matrix in bbmpeg ?
:arrow: If I follow you correctly, if you do not do that, all the KVCD muxed with bbmpeg will be unscramble with the wrong key :!:

Okay, that was the point where I didn't join the pieces. Now I understand.

BTW, I can't install it on my Athlon ! I have big issues with MSVCR70.dll that is present but ffvfw installer complains about it :-(

Is there someone that have an idea ?

:? Tell me more ...

I have Athlon and its instaling without problems. But I have had this problem with ffdshow few months ago on clean (without any instaled codec) machines. I just downloaded missing file and put it to winnt\system32 (maybe it needs to be regsvr32 'ed ;) )

:? Tell me more ...[/quote]
Look at into bbmpeg, just under the part where you fill the name of the video stream and audio stream you have a place to put the files for the 2 matrix.
According to what Inc just said, if you do not put somethign here, your image will be ugly.

I'm waiting for an answer from him :-)

What do you think I did ? (I have two PC linked in network -:)). But it's not enought ! BTW they were already there as I told. And there is nowhere to find an INSTALLER for this sh!t (M$ product for sure !)

The builds just before the 2004 ones don't need the msvcrt files and they are based on the same cvs so don't know if you would see much difference anyways.

http://www.ligh.de/software/ffdshow/...w-20031128.exe

http://www.ligh.de/software/ffvfw/ffvfw-20031117.exe

Don't yell at me :bawl:
The way I proposed work for me. I even create modified instaler for other people, and they have never declare problems. Any complaint to the Bi-G :lol:

Thank oyu Dano, I will see.

But in fact I have two problems : ffdshow does not install at all (but I do not need it actually). But ffvfw installs itself once (the second install fails because of the dll) but... I DON'T HAVE IT IN VDUB ! :-(

EDIT: OKAY ! With release given by Dano it works. But is the encoder really the same quality ? Inc, which was the version you used and taht gave problems to you ?

Hi Phil,
Inc is online but probably doing something else. :?
Though I don't know for sure which build he was using when he had problems,
I can assure you I had issues with the one Dano has just told you to use.
I'm using iNTEL cpu and I didn't have any problems, so I went to check on an AMD cpu and I also didn't find any problems installing it.
Any chance you're running a French version of WinXP?
Try checking with a friend who is also running French WinXP.
I say this because I've had some issues with Portuguese WinXX so many times that now I only use English versions :evil:
C ya

It seems the problem is deeper : I can't even install half of the patch provided on windows update ! The install fails and for sure this "wonderfull" M$ crap don't even tell why :-(

@Phil
We sure could be running something stronger on our PCs nowadays.
I mean WinXP is nice and all but it's still too dumm.
Suddenly when we least expect it, we start seeing the OS failing because of some rather buggy
application we installed last week...
That's just something the OS should try to solve by itself...
It should already have some repair routines like we see on some not as good looking OSes...
Worst: It's got own willing :evil:
Good luck Phil. :?

In fact I just foudn the problem : that was a service that I stopped (ie desactivated) days before and was mandatory :-(

Friends I had a very nice time with an ugly bad converted PAL capture where someone before did a conversion from 23.976 to 25.000 which has ben send via cable :arrow: Blending, interlacing and everything which gives headache!.............. :evil: THIS SUCKS! 2 Hours to solve that using Selectevery() etc.

So Phil,

Yep you're right, our encodings would come out ugly IF the matrix wouldn't be placed within the video mpeg, as I did read in the www.

If you open Restream you can see that its not only the VOB or the mpeg which can be opened, its also m2v m1v source etc.
That means as said the matrix values for descrambling are within the encoded videostream.

Its nice, just when you did load that video stream, move with the mouse over the little floppy disk "safe" symbol in the matrix sections and a pop up will appear which hows you all the values.

By this you can get the orig Matrix where your orig DVD has been encoded with.

;-)

@ Kwag

You're right, as I forgot that the ffvfw matrix is only 8bit in its values, but we still can use that matrix
By copying the values into the matrix section of ffvfw, just replacing all the 99 with 64 and add in the avisynthscript at the end a

DCTfilterD(4)

At least at the intra section this will all 4 last diagonals cut (like 99) and in the non intra section, well only one diagonal will be still used with lower cutting parameters, but by this we can try.

That's what I thought (I never understood why bbmpeg allow to give matrix files. BTW, I do not use bbmpeg :-))

That is a new answer for the guy on other thread taht was asking "how to distinghuish regular MPEG from a KVCD one" :-)

Where? The mpeg4 HVS matrix comparison? 8O

Lets cut da frequencies! :cut:

:D

Fortunaly for newbies, while you are all testing ffvfw, I'm still reading the forum and helping them :nose:

http://www.kvcd.net/forum/viewtopic....910&highlight=

:lol:

We should add to our signature ....

"away in *ffvfw headache holidays* will be back soon .... :drink:


Uops:
It was the build from Nov.2003! Since I updated to both (ffdshow/ffvfw)latest builds (I think they are the Jan.2004 ones) the big problems where gone.

That is what I thought. Fortunally I managed to have the 2004 version to be installed and working.

I've got the same problem with the 2004 version, you said in a previous post that a service was mandatory, which service did you reactivate?

Anerboda

Can't say ! I reactivate all services that weren't to see "if it fixes something". And as it did, I can't tell wich one was the problem :-)

But, be sure first to have the dll msvcr70 and msvcp70 in the system32 directory of windows.

Thanks Phil, actually I didn't have the DLL's in my System32 folder, so I found them at first and placed them in the folder, and then it worked, thanks again.

Next I would like to know if anybody have compared the mpeg1's(KVCD's) produced by TMPGEnc to the mpeg2's (SKVCD, not DVD) from ffvfw, I mean m2v with a low enough bitrate to fit in one 80 min CDR.
I know ffvfw is considered superb in encoding DVD's, but has anyone produced SKVCD's (with notch-matrix) using ffvfw and compared these to KVCD produced by TMPGEnc?

I have read this thread (all 17 pages) 5 times, and maybe I've missed something, but I didn't find an answer to clearly remove my doubts.

Right now I'm doing some tests myself. I have encoded a 1:30 min sample in TMPGEnc targetting KVCD, and the same sample in ffvfw for SKVCD, both resolutions 480x576. In TMPGEnc I used a CQ of 65 which gave me a sanplesize of 14000 KB (too big), and in ffvfw I used one pass CBR with a bitrate of 1095 (average bitrate according to CalcuMatic), this sample was 11920 KB (very close to wanted samplesize).Used the Notch-matrix in both encoders.
Visually I didn't see any difference (I was looking at the encoded samples in VDubMod).
I hope that setting the CBR to average bitrate will hit the target of 750000 KB.....Doing a full encode right now....
3 hours and 20 minutes to go....

Anerboda

You can actually go a little (~200kb) over the Average Bitrate in CBR mode of ffvfw. This it what I experienced. Tell us your results! 8)

I did test, and I do continue to make my KVCD with tmpgenc.

All the thread is talking about KDVD, taht's why you didn't find your answer.

The only problem is that the settings you put does not seems to have any usage !
Yesterday I did test with 8000 and 5000, thay both gave me the same sample size, and the peak in each one was over 8000 !
I did after that an encode with 2500, the file size was half it was previously but the peak was near 5000 !

If you plan to make KVCD with this, you can't be sure that you won't have a peak far above the limits allowed on a VCD (audio+video = 27xx (do not remember exactly)).


Anerboda[/quote]

I didn't plan to make KVCD, but SKVCD, don't know if my standalone will accept peaks higher than 2720 or something like that.
But I viewed the m2v stream from ffvfw in bitrateviewer, and it showed peaks up till ~1800kbit/s and an average of 1045kbit/s, so maybe I could have raised the CBR to 1150kbit/s in ffvfw??
Waiting for my full encode to finish... :D

Anerboda

That is just playing with words. I guess that what you call SKVCD are 544*576 MPEG2 KVCD (in other words MPEG2 KVCDx3). All are KVCD.

Moreover, The word "SKVCD" does not, and never did, exist. Go to this page and see :

http://www.kvcd.net/dvd-models.html

You have KSVCD but no SKVCD.

It won't.

Hmm... I just stumbled on another bug I think. In CBR mode, ffvfw does not seem to encode as 16:9 output, although it is set. BV says 4:3, and also PowerDVD plays it as 4:3. Well not a big deal, I can patch it via DVDPatcher, and then it works. :roll:

I know, I just spelled it wrong, sorry. What I meant was a m2v stream in a resolution of 480x576.
My standalone plays perfectly KVCD's at 544x576, I was just curious if KSVCD at 480x576 encoded with ffvfw looked better...and perhaps if prediction with ffvfw was more accurate...
But I'll test some more...
And when I get a DVD-burner, I will try ffvfw with .m2v-streams in 704x576 resolution, until then I'll probably stay with TMPGEnc and KVCD.

Anerboda

Jell, ... right thought ;-)

According to the encoding itself 16:9 or 4:3 means nothing to the image encoding way! Its just a flag where the decoder sees how to get the final resize/AR so it fits on Tv!

So its not important what ffvfw does according to AR, you just can set this afterwards using DVD Patcher.

If you want to encode anamorph, its just a resizing thing in i.e. Avisynth that the image itself will be squeezed (anamorph) 704x576 encoded ... and adding the AR flag afterwards. ;-)

When using lower resolutions, like 352x480(576) or 480x480(576), you won't get peaks of ~8,000.
They will be arounf half (~4,000).
You might want to use the notch matrix at these resolutions, and set your MIN and MAX Q to a value of 2.
Even set like this, the file size will be about half the size, as when you use the standard matrix, but the quality will be about the same because of the reduced resolution.
Yesterday, I did just that, and this is what you get:


http://www.digitalfaq.com/archives/i.../2004/02/4.png

So at that resolution, with ffvfw doing MPEG-2, we can do KVCDs. :)
At higher resolutions, we must use the internal matrix, which is fine for DVDs.
Or we can do KDVDs at half D-1 352x480(576) with the notch matrix, and it will look like the screenshot above.
Not as sharp as 704x, but still a blockless picture, even on high action scenes.

-kwag

It's the other way around Phil :)
It's SKVCD

-kwag

Arghhh.... :jawdrop:

:imstupid:














(if you are looking for me, I'm already out) :-)

Hi Anerboda.
AFAIK rds_correia's point (and mine to), when he begin "Mencoder for windows" post, was to produce KVCD's with something faster than TMPG (CCE). When Dano pointed at ffvfw, I was only pointing in KVCD (MPEG1 or 2). I used ffvfw some time ago for Theora test, but i never realized that is a powerfull MPEG2 encoder.
I did some MPEG2 samples, but bitrate goes as high as ~4000 and above. I get a lot of underflows in bb, when muxing audio and video, but none without "Add SVCD scan offsets" and with "Align sequence headers". Files muxed this way are played perfect on Panasonic NV-VP30, but this baby eats everything alive 8) I'm no specialist in this settings, so I can't say what they realy do. Some people here knows what for are they, ask them ;) I now make full lenght 480x576 move and will se results tomorrow.

PS: I use quality mode, while CBR mode is useles for low bitrates :(

This post starts giving me headache... I'm lost. I've been trying ffvfw with 'constant bitrate' and the parameters from kwag but rising or lowering the Max quantizer difference doesn't makes the file bigger or smaller, it remains the same size.

But anyway, changing the bitrate and the GOP size (I had to set it to 23, 24 isn't accepted) I have done a 1h28m 4:3 movie in SKVCD ;) that looks great and is played perfect in my player.

My question is... Where can I find any information on how quantizers affect quality? I want to change those values an try to find good ones or at least some that fits my needs.

And thanks for having discovered this codec, doing KDVDs is incredible!!! (and without size prediction as Kwag said) :P

I'm looking forward for a guide and an optimal script.

KDVD now a reality :)
 

Good news, and new finding :D

The problem with the matrix, and ffvfw, has been solved :!:
:ole:

Set the KBPS Bitrate slider to 6426
This will fix the VBV buffer size to 112, as shown in Bitrate viewer, and now bitrates will exceed 7,000Kbps with KVCD's notch matrix :mrgreen:
All other parameters are set as in my previous post, with the last screenshots.

Now here's what you get with KVCD's notch matrix:

http://www.kvcd.net/ffvfw-notch-kdvd.m2v

If you encode a fast action sample, you'll see bitrates beyonf 7,000Kbps.
The sample above only goes up to around 2,000Kbps, because it's low action.

Enjoy :mrgreen:
-kwag

Note: I'm calculating the correct bitrate value, to get the correct size for KVCD's :cool: which I think will now be possible with this correction.

Hi: Kwag, can you make a thread only for ffvfw KVCD and KDVD optimal config?
In this one, is to hard follow up all the changes.

Sure will :cool:
Let me get some screenshots and I'll start a new thread.

Edit: BTW, for KVCD, the correct value to get a VBV buffer of 20 ( actually 40KB) as seen in Bitrate Viewer, is 1147 ;)
I'll explain in the new thread.

-kwag

@ Kwag
8O NICE! But the Bitrateslider (as I figured for myself now out) was up to now the only adjuster I could lower and higher the final filesize! Uops!
Cause the way you explained in your last pics didn't get my "prediction" working ;-)

But the Q curve of Bitrateviewer now in PAL ist straight and linear!!!

But I will look further!

PS: In that new CBR Mode the FPS encoding speed index decreases about 30% 8O

Did you also check in the left right corner of your task bar, that little FFvfw tab!! There you can see the graph while encoding and see how much the quantisation will take!!!

NICE! By this I can now directly test the optimal qantisizing for encoding! Extreme coool!

Nope. Doesn't work that way. That was the "illusion we got, because changing the VBV buffer size, would also change the file size ;) Yep. You must change the quantizer values to get a file size change, and leave the bitrate slider set to the correct values shown, so the VBV value is constant for the required encode.

-kwag