Re: Kwag
 

Be carrefull to not mistaken ffdshow for ffvfw. The fist one is a directshow filter (used for playing only) and fir sure you won't do any encoding with that :-)

Hi Karl,

I gave it a shot, but this time using a different video, a XVid documentary. I'll try to check if sampler() prediction is working with FFVFW. My target is 6.4Mb.
Using max quantisizer 25 and 2 B-Frames Quality 100:
No noise gave me 6.2 Mb
Luminance Noise Strenght 6: 6.3 Mb
Luminance Noise Strenght 7: 6.7 Mb
Luminance Noise Strenght 8: 7.2 Mb
The growing is not linear, but it's didn't get as big as my last try with the LawnMover Man. Maybe it has something to do with the fact that the Lawnmover man DVD is very noisy... :!:
I'll encode using Luminance Noise Strenght 7 (don't mind overburning if needed :D ) and let you know if prediction worked as it should.

Hey Phil, ffdshow filters can be used for encoding in ffvfw. You need to install one of the more recent builds of ffdshow and select "Enable Image Processing" in ffvfw. This only affects ffdshow encoding options, if you open the decoding options in the configuration dialog from ffdshow you will see that none of the changes you make in ffvfw are there. As a matter of fact there is an entry in VirtualVCR for "ffdshow encoder", separate from ffvfw.

Prediction went way off... :(
I needed a sample of 6.4 Mb. Used a 6.7 Mb, expecting the full documetary to be 600 Mb. I ended up with a 491 Mb file. I don't know if this has to do with the fact that the documentary is very compressible (lots of almost static people talking in front of a static background in most of it), but maybe we need new prediction methods... :?

EDIT: Prediction is Working, read my next post!

Or simply, no prediction at all :D
Encoding as constant quality ( just like the Panasonic DMR machines do), where it's guaranteed X quality on X space.
That's the way I'm doing it, and for example, I'm currently burning a KDVD which contains Red Planet (107 minutes) and Count of Monte Cristo (127) minutes, and I had 1.4GB of free space left, enough for one more movie :)
But I just had to burn this two movies, just to see some parts on my HDTV, and see how they look :cool:
The way I see it, with wide screen movies, we can easily fit three 2 hour movies on a DVD with full DVD quality. And this is better quality than what we got with TMPEG doing KDVD with a MAX bitrate of 5,000Kbps.
I see peaks of 7,000+ Kbps on the Count of Monte Cristo movie (where I added the noise). Still within the DVD specifications.
This can only get better :!:

I can't get enough of this encoder :mrgreen: :angel: :drink:

-kwag

Well, I am not sure if I understood, but there's still no way to know if any three movies combo will fit, right? I understand that you're saying that in your tests till now this have been happening, but I don't get how this can work since compressibility varies from one movie to another... :?

I am, like you, really impressed by the quality of this encoder. The only thing missing is the ability to control bitrates... Or maybe it's the freedom to use any bitrate that makes it so good... Who knows :!:

EDIT: The documentary encode was done at the speed 0.36 realtime (took 2:30 to encode a 56 min documentary). I tried to encode the first 10% of it using CCE and went up to 0.47 realtime. I guess the result I had previously is because CCE starts slow and gets faster with time, since I was encoding a small sample it never got faster enough. Anyway, in my machine, CCE is only 30% faster than FFVFW with noise actived. But FFVFW seems to produce same or better quality and it's free.

EDIT2:Preditcion is working! I made a really great confusion! My target, with sound, was 600 Mb! The video was supposed to be 517 Mb, so 491 Mb is good considering that it's a very compressible matherial! That's great! :D :D

Here are my experiences with this encoder. I made a script with resize (704x576), temporalsoften, undot, stmedianfilter and mergechroma(blur(1.5)).

Refering to encoding time and file size:

Encoder - Time - File Size
ffvfw - 2:06 - 5842 -> (kwag's parameters, gop 15)
ffvfw (blockb)- NA - 5882 -> (method="noise",detail_min=1,detail_max=10,varianc e=0.3,seed=5623)
ffvfw (noise 5)- NA - 6593
CCE- 1:19 - 5869 -> (Q 40, 16:9)
TMPGenc- 2:01 - 6618 -> (KDVD template, CQ 70, 16:9)

I compared all the samples in a scene where a starcraft explodes. And, to my eyes, in this particular scene, CCE is the less blocky, closely followed by ffvfw with noise (I begun with 8, but the file grew a lot, and didn´t look well).
If you tell me how, I can post some captures.

BTW: I selected in ffvfw a 16:9 aspect ratio in output, but when I tested it later, was encoded in 4:3 aspect ratio. And this happens in every test I did. Am I doing anything wrong?.

I'm processing an NTSC anamorphic Widescreen 235:1 DVD with ffvfw with Kwag's settings (at first of this thread). I wish to produce an NTSC anamorphic Widescreen 235:1 DVD. I will use Forcefilm in DVD2AVI. My avs only contains the mpeg2source line with the d2v file.

:?: Which settings, if any, should I change in ffvfw from Kwag's settings?

I will run Pulldown.exe at the command prompt as: "pulldown myfilmIN.m2v myfilmOUT.m2v".

:?: What settings should I choose in DVDPatcher? The source is 720x480 and the destination will also be 720x480.

With the tries so far, I can mux with mplex, but when I try TMPGEnc Author, I get too large GOP size error [NTSC DVD requires 18]. :?: Any way to fix this?

:D Even though I'm having some difficulties, the results look very good far. Just need to learn how to work this thing!

Hi Nick,

Set the bitrate to 8,000 which is within the DVD standard.
Remember to just patch the header. You must set the GOP to 18 :!:
Maximum I frame interval = 18 Me too :lol:

-kwag

wooooow... seems as if I have missed lots of exiting new stuff :D

I drifted away into mpeg4 as I bought my self an mpeg4 standalone :D

But as for 2 audio streams I still need mpeg1/2, I guess I'll have to do some reading on the forums :D

Well, again, if you think it's of interest, and tell me how, I can post the screenshots of tests I did (I posted this before forum update, so maybe you didn't notice it). I don't have a web page to upload them.
About the 16:9 issue, I don't know if I did anything wrong before, but in my last test I got 16:9 aspect ratio.
Have you tested Full processing against Fast recompress modes?. Is it faster?. Does the visual quality suffer?. I think we have to improve the encoding speed, don't know if it's a question of change settings.
In 720x576 size, a noise level of 7 or 8 looks too noisy, don't you think so?. Or is it just a question of high resolution?.
Finally, I processed with TMPG DVD author without patching header, and got a warning about high bitrate, but could go on with the process. Didn't find any problem playing it with PowerDVD. Is it really necessary to patch the .m2v file?
You see, lots of us interested in what ffvfw can do... No answer from the developer yet, I'm afraid, isn't it?.

Kwag, settings...........
 

I've only changed the ffvfw settings to the one's shown in your example, including the changes to "Noise". :?: Since this is anamorphic widescreen, should I leave everything at 4:3 in ffvfw and DVDPatcher ( not 16:9) ?

Just did my second encode, a DivX conversion of "007 - Dr. No". Prediction worked fine: target was 707 Mb, got 718 Mb. This add noise is amazing for DivX conversion! Using the latest Dialhot script plus Noise 8, those annoying DCTs in the source all become noise! Fantatisc picture, about the closest I got to a DVD. I can't believe this free encoder can do a better job than CCE and still be very fast! :D :D :D
Kwag, you're right, it can only get better! :D

When using DVDPatcher on the m2v from ffvfw (after pulldown.exe), do I set the framerate at 29.97?

This is NTSC forcefilm from dvd2avi.

Forgive all the questions, but ...............

Can you say which is your prediction metod? Thank you.

Ok, I'll try to outline the basic steps. It's like my CCE method adapted to this encoder.
1) Download my calculator here: http://www.jltoca.uaivip.com.br/files/KDVDCalc.exe
2) Open your script in VirtualDub and check movie framerate and movie lenght, in minutes.
3) Open my calculator. Select CCE KDVD Preset. On initial Info, put the same GOP you intend to use in your KDVD. GOP is called "Maximmum I frame interval in this encoder, you can set in the Generic Tab of the encoder. I suggest you use 18 for NTSC or 15 for PAL, to be within the standard.
4) Put the correct framerate for your movies, should be 23.976 or 25 fps
5) On movie caracteristics, put the movie lenght in minutes and the audio bitrate you're willing to use, leave audio size in black. Do this for each one of the 3 or 4 movies you want to fit.
6) Click on calculate. Pay attention to "Desired Sample Size", it shows what is the size of your ideal sample.
7) Get sampler.dll plugin for avisynth and add this line to the end of your script: sampler(length=MAX_I_FRAME_INTERVAL), Replacing MAX_I_FRAME_INTERVAL by your GOP.
8 ) Encode the three scripts with sampler, and check the size of the produced MPG. If the sizes (combined) match the one you got from the calculator, you can take out the sampler line and do the full encode.
If it doesn't, try to vary encode parameters: use less noise (6 or 7 instead of 8 ) and/or quality less than 100. But remember that with this encoder quality drops very fast with the quality setting, so don't put it too low. Encode again till you get the sample size calculed before.

That's it, I hope it has helped. Looks a bit complicated because I tried to explain every step, but it's very simple, and if you have someday followed my CCE guide you'll see that's exactly the same thing.

Hi vmesquita:
do correct me if I'm wrong. I tried your nice calculator, and it seems to me that the sample it calculates is about 1% of the film length. It would be nice to have the option (set to 1% by default) to change this percentage, if we want a bigger sample.
And, yes, I also find this encoder very promising, we just need to know how to adjust parameters (bitrate for instance, or how to get better speed), and it will be THE encoder :lol: .

Hi digitall.doc,

The samples have to be generated by taking slices that have the exact GOP length for the prediction to be the most precise possible. But you can try the following: use the GOP parameter in the calculator and in the sampler() line as the GOPxY of the actual GOP, where Y is a integer. So you could use 36 or 72 for NTSC, just remember to keep the actual GOP as 18 in the encoder and always use the same values in the calculator and the sampler() line. Maybe this would lead to a more accurate prediction, but I never tested.

THIS Sucks!

I did a full encode again on Truman show with the latest build of ffvfw!
Quality now is better!! (it was the build, released before, that gave me a headache!)

But even when I demultiplex it and assign a conform max. Bitrate .... when this stream loads into Bitrateviewer, almost at the end of the loading process, Bitrateviewer BOMBS!

I chacked that stream using Restream:

FIrst it gave me an information that thius mpg2 stream gots a nonlinear quantisation! Which is normally a spec. of mpeg1.

And now here it comes: A VBV Buffer of 7!!!
And thats why I cant multiplex it for example in BBmpg where a lot of underflows do occur. I did that encoding using 352x576 and a file came out which fits incl. Audio exactl. one CDR80.

But ... to me it seems that there is no VBV Buffer to set in ffvfw!

:cry:

I agree. By using the latest build of ffvfw from 2004 things got better although I never ran into the problems you did.
As I said never got bombed by bitrateviewer when I demultiplexed with tmpgenc...
No multiplexing here by now because I'm still getting to know the encoder. Can anyone confirm this info by Inc? I'm trying it tonight also.
Yes indeed it seams there is no way to set the vbv buffer and also there is no way to limit min and MAX bitrates...
This is very sad, because it really seams to be getting better using a static script instead of MA.
This way encoding speed leveled with tmpg's and quality is still superior only inferior than cce that is worth big $/€ :twisted:

@Kwag
Anyway I will convince you to build a WAP version of the forum :?:
Since I'm on the move for the next weeks it's going to be hard to keep up to date because
I won't have broadband Internet access.
But I'll still have my mobile phone :twisted:
Cheers

ffdshow is the decoding part only of the codec (coder/decoder) called ffvfw :!:
I never said they are tied together, they are TWO COMPLETLY SEPARATED PROGRAMS using the same code. But one is used by the system as a directshowfilter (and can't be used by avisynth for instance if you don't use directshowsource), the second one is referenced as a codec under VFW API of windows.

The same way you can find directshow filter part only of Xvid codec. If oyu isntall only this part, you won't be abble to do any xvid with avisynth as it will said "no codec found for this source".

One more time CODEC and DSHOW FILTER is not the same thing

That's strange. It hasn't bombed on me, not even once :roll: Haven't had a problem with that either. I don't get any under or over flows.No, there is not. However, if you can find one of the early beta version of TMPEG, TMPGEnc-0.11.20.97.zip with the english patch, there's an option in the MPEG tools that will let you correct the VBV buffer size. Give that a try.

-kwag

@All
Don't seem to be following you guys (Kwag/Inc).
Is there a problem with a vbv buffer being so low or not?
Because I should be experiencing muxing problems also (since I'm PAL also) and from a small 5 minutes
clip I did today with ffvfw and headac3he muxing with both mplex and bbmpeg gave me no
under or over flow errors at all.
If I wouldn't want to use VdubMod would there be any other app I could use instead?
@Kwag / Hydeus
Any news from Milan yet?
Do you still need to talk with him?
Would documentation help us here?
I hate :evil: when somebody suddenly disapears from the net...like Muaddib and Jorel...
Cheers

Hi all,
hey incredible, I also didn't run in those problems ever. I don't know if it's a matter of size (you said 352x576). Maybe this codec, the less the resolution, the less the max bitrate and less VBV... do you think this is possible?.
I did my tests at least with 528x576 res., and could multiplex with bbMPEG without a trouble (no over nor under flows). Even I fed once TMPGEnc DVD author with a MPEG2 file without patching, and just got a warning, but did create the IFO and VOB files, that played fine in PowerDVD...
I guess that we have to test, test, test, and try different settings (what about changing in Motion Estimation, EPZS estimation to Adaptive?) in order to get the best of this codec. As I don't have idea how to adjust it, I just wait for any suggestion to test it.

I did'nt disapear (few free time, alot of work :evil: ). I also have no reply from Milan, but manual is no use for now, however few bugs need to be squash.

@Incredible
I also have this underflow problem with BBmpg in mux, and this files are jerky playable on standalone (when recorded with VCDEasy). Also the same files burned with Nero are perfect. Any idea why??

And for another thing. When setting postprocessing to full in ffdshow (under input/enable image processing) gave me smaler file with the same quality. Please anybody confim or deny. My eyes are subjective.

Edit: On the other hand, encoding is almost twice slower :(
Edit2: With postpr. and noise, speed increes a litle, but image is litle blured.

@Phil,

Sorry I didn't mean to imply that ffdshow was a decoder, just thought you didn't realize ffdshow's preprocessing filters could be accessed ffrom within ffvfw. I see I was wrong.

One question for Kwag or others : I just did a little test with ffvfw and in the end my video doesn't have a constant Q like yours in bitrateviewer.

The value runs from 2.25 to 3.36. Is that because I do not use pulldown (I am doing PAL) ? Or what can be the settings I missed ?

Same in my case! Strange!

You would ask now why my Bitrateviewer doesn't bomb, because in case of small samples it does import it well, but not full long ancoded movies.

http://www.digitalfaq.com/archives/error.gif

Watch the VBV Buffer size and also - although mpeg2 encoding - the (in regular to mpeg1 allocated) LINEAR Quantisation scaling.
(same output when demuxing before!)

Same thing here, and it's not a PAL issue since I am doing NTSC. But picture quality is still great.

Except than my Q curve is more like the yellow one on your snaphop ! In fact both lines look like a montain lansdcape :-)

I just have to reinstall my bitrateviwer as it hung as soon as I click on "open" icon :-(

Mine is also 7 and on the snapshop provided bu Kwag, it was also 7 :!: So that not seem to be a problem.

Hi: Why the Maximmum I frame interval in usual KVCD Template is 24 or 25 (ntsc or pal), and for ffvfw is 18 or 15?

Well my Q curve does look like influated by an intervall, so where do we have an intervall in our encoding: In the max. I framecount and also within the GOP itself. And I think its a issue of different B and P Frame Quantisation :idea: :idea:, even if its set equal in ffvfw settings.

Maybe that VBV 7 issue will give no Problem on DVD Mode or on diff. Standalones (where in regular 224 should be the one) but when muxing with bbmpg I got many many underflows as one encoding I did, did fit amazing on one CD80 and thats why I tried to mux a SVCD VBR using that.

Cause we're doing in here kDVD encodings and thats why we use 18 NTSC and 15 PAL

I know that ! But I wanted to say that Kwag also have a VBV=7 and DON'T have any underflow ! So that is not the VBV that cause this.

Because we're testing ffvfw for making KDVDs. Not KVCDs.

Edit: :oops: was already answered :lol:

-kwag

Yes this is true Phil. The three movies I've encoded, all have a VBV size of 7.
But all have been progressive 23.976, and then ran through pulldown.exe and then DVDPatcher to patch the header to 8,000Kbps.
When muxed with Mplex, I've set "~DVD" in the muxing option, with all default values.
Multiplexing is flawless, without any underrun or overrruns.

-kwag

There really seems to be a PAL issue 8O
I did some comparisons between 23.976 and 25.000 using ffvfw, cce and TmpgEnc.

Ok, the Pal stream is a bit different in its treatment cause I did set the Slicer to NTSC and PAL GOP specs, therefore the sliced samples where shurely different, but that should not influate the average behavior of the modes I saw as posted here:


http://www.digitalfaq.com/archives/error.gif

But the very worse Q curve in compar. to CCE and Tmpg stands in no relation to the real quality in the video.
But Anyhow ... at the same endfilesize, ffvfw does look a bit better if seen as a whole! But if we would trust totally in bitrateviewer, Tmpg should look 3x worse but it doesn't if seen in the pics below.
(rised gamma/Luma via levels afterwards to compare dark parts)

http://www.digitalfaq.com/archives/error.gif

@incredible
This ffvfw screenshot look amazing... 8O What was the blockbuster setting you used? My encodes with ffvfw don't show this straight Q curve... :? I'll post later. :wink:

The pics are 2times scaled! And enlightned by Levels() But I think you did checked this already ;-)

This Setting ONLY affects the plain parts and NOT the details/Edges.
You can play with the "variance" values which do affect the noise level.

Also apply that one to your script at the very end before adding the blockbuster and then the borders:

(Only to use at 704,XXX encodings cause they wont be stretched horizontally that much on Tv!!)

Which would mean if seen in the Matrix (here now in TmpgENc)

http://www.digitalfaq.com/archives/error.gif

That gives in my 1% Sample an advantage of compression from 13.886 to 12.898 kbyte Means 100mb less on the whole final encoding, but that also depends on the movie source and its detail quality.

On the Tv later you WON't see that less sharpenes as it even does not affect the sharpeness that much as compared to our MergeLumaBlur commands in the Ma! But its not recommendable when encoding at for example half DVD size (352x480).

Hi Inc,

Are you aware that DCTFilter is included in ffvfw :?: ;)

http://www.digitalfaq.com/archives/i.../2004/02/5.jpg

-kwag

YEP! But not DCTfilterD() ! :mrgreen:
Without that, only the 2 right collumns would be filtered, and so I did the job as a whole in AVS.

And ... I like scripting :wink: