KVCD VBR??

[gonzopdx]

I post this here because apparently the KVCD forums are a POS and every time I try to post, the f'cking site goes down.

Alright, so I've done QUITE a bit of video encoding using TMPGEnc and I wanted to give the KVCD templates a shot.

I was going to try out KVCDx2 (or should I try x3?) to encode from a DVD source..

Personally, I think that fiddling with CQ settings is a HUGE pain in the ass and I really don't prefer to sit through a 6-12 hour encoding session just to find the file is too big, then have to sit through a 6-12 hour encoding session AGAIN just to get the right size..

So my question is this: would using a 2-Pass VBR (to get the file size perfect on the first shot) work with these templates? From the description on the KVCD page, these templates are just modifications to the GOP and Quantize matrix.. so I don't see why VBR wouldn't work.. or am I missing something entirely?

Please respond soon as I want to try this ASAP =)

TIA

-G

[hwoodwar]

I use the SKVCD NTSC film template for CVD all-the-time with 2-Pass VBR. I've changed it so it no longer does the inverse telecine, so thereby the framerate is 29.97fps, use high or highest motion search precision and on the 2-Pass VBR usually use about 1000 min, 2200 avg, 4000 max.

Note: all three of my standalone DVD players can handle over 5000kbits video bitrate for a XSVCD.

I use the "Project Wizard" so I can adjust the average bitrate to fill the CD. IT does pretty good with it's estimates.

I am not encoding from DVD. I am capturing from VHS tape, or TV card tuner. Right now I'm encoding the mini-series "Taken".

I did try modifying the x3 (MPEG-2) template once to CVD, but in my tests I liked the video quality a little better of the modified SKVCD NTSC Film.

[kwag]

I post this here because apparently the KVCD forums are a POS and every time I try to post, the f'cking site goes down.

Could you be a little more specific? KVCD.Net has been up 99.9999% since it started. Maybe your internet provider has some routing problems ?

I was going to try out KVCDx2 (or should I try x3?) to encode from a DVD source..

I almost never use the x2's anymore. The x3 is the perfect balance of resolution/bitrate for just about any movie. I only use the x3 for targeting 2 CD's. You can fit almost every 2+ hour movie in 2 CD's.

Personally, I think that fiddling with CQ settings is a HUGE pain in the ass and I really don't prefer to sit through a 6-12 hour encoding session just to find the file is too big, then have to sit through a 6-12 hour encoding session AGAIN just to get the right size..

You don't have to fiddle with guessing anymore. Go here: http://www.kvcd.net/forum/viewtopic.php?t=68 and read Q-9 and the links under Q-9.

So my question is this: would using a 2-Pass VBR (to get the file size perfect on the first shot) work with these templates? From the description on the KVCD page, these templates are just modifications to the GOP and Quantize matrix.. so I don't see why VBR wouldn't work.. or am I missing something entirely?

Once you use file size prediction ( of KVCDPredictor program ) you'll think twice about X-Pass encoding

-kwag

[gonzopdx]

Could you be a little more specific? KVCD.Net has been up 99.9999% since it started. Maybe your internet provider has some routing problems?

<shrug> every time I tried to post to the forum, I would be unable to return to the site for five to ten minutes, and when I finally could return to it, my message would not be posted.

You don't have to fiddle with guessing anymore. Go here: http://www.kvcd.net/forum/viewtopic.php?t=68 and read Q-9 and the links under Q-9.

Once you use file size prediction ( of KVCDPredictor program ) you'll think twice about X-Pass encoding

Alright, I got the KVCD predictor, but what am I supposed to do with it? This gives me a sample size -- what am I supposed to do with that? I don't understand the relationship to the Scale Factor and Sample Size and the final movie size. I read the threads and don't find any explanation. Am I retarded?

As an example, I'm trying to encode two 25 minute clips to 1 80 minute CDR. I put in 25 mins in the source length, put in the size of the audio (28 megs), use the default Scale Factor of 0.95 and it gives me a Sample Size of 54.03 megs. Wonderful. I double the inputs as it's two clips (ie: 50 minutes of video, 56 megs of audio). Again using 0.95 for the Scale Factor, Sample Size of 26.03 megs. Fantastic. What am I supposed to do with that number? I up the Scale Factor, Sample Size goes down. Down the Scale Factor, Sample Size goes up. Neat, the wonders of division in effect. So what am I supposed to do with these numbers? I don't understand how this will help me figure out the Quality factor in TMPGEnc and bitrate to determine the final size of my file.

Please explain.. Thanks.

EDIT: I took the number of frames for one clip (35240), doubled it, divided by framerate (23.976) and 100 to get my "magic number" (call it that as I'm not sure what it's for -- I understand dividing by framerate to get time, but why divide by 100?). I multiplied my sample size by the "magic number" and Scale Factor and got a file size that's too small. I upped it up to 1.00 (I'm assuming this is for 100%) and still get a file size that's too small for my target. Does that mean I can use CQ at 100% and 2500 for the bitrate..?

[SansGrip]

Alright, I got the KVCD predictor, but what am I supposed to do with it?

KVCDP is intended to be used in conjunction with a minor modification to your Avisynth script. The modification selects 100 1-second samples over the course of the clip which you then encode with TMPGEnc at a given CQ_VBR. If the resultant file size is too big, reduce the CQ_VBR -- and vice versa. You can read all the details at kwag's forum. I believe I rather illogically started the KVCDP thread in the Avisynth forum.

So yes, you're still monkeying around with the CQ_VBR somewhat, but 100 seconds is a lot quicker to encode than an entire clip. Once you've got the correct CQ_VBR (which with a little practice only takes two or three iteratrions) the final encode then takes half as long as 2-pass VBR and is about as accurate.

Besides that, it sounds to me like you're not using the latest version (0.2), which includes a wizard-type interface to help decide the correct CQ_VBR for a particular clip.

Regards,
SansGrip

[gonzopdx]

Where can I get 0.2? I just got it from the kvcd.net page -- I believe it links to 0.1b

I tried using the FitCD method and using the file size prediction option and followed the instructions in the FAQ exactly, but when I load the AviSynth script into TMPGEnc, it tells me it's going to take two hours+ to encode, and it ends up encoding the whole movie, (or that's what it looks like).

On another note.. I am still playing around with it. I decided to encode a 30-second high-action clip from what I want to encode with different settings to see what kinds of results I'd get.

I used the VBR calc available on this site to get a VBR number (2005 kpbs). I plugged this into CBR to see what size file I'd get. The resulting sample clip is MUCH closer to my actual target. I also divided 2005 by 2500 to get 80, which I put into CQ for Quality (80%) and also got a clip very close to my actual target size (though CBR was way closer, EDIT: but CQ looked better).

How do I determine the bitrate to use? It seems that these tools and faqs help determine the Quality factor, but how do you know what bitrate to use? Should I just keep it at the max and change the quality?

Thanks..

[kwag]

Hi gonzopdx,

Here's the KVCD Predictor thread: http://www.kvcd.net/forum/viewtopic.php?t=1729

Methods and techniques are explained there.

-kwag

[gonzopdx]

Hey kwag..

Hi gonzopdx,

Here's the KVCD Predictor thread: http://www.kvcd.net/forum/viewtopic.php?t=1729

Methods and techniques are explained there.

-kwag

Thanks, but..

I read the threads and don't find any explanation.

EDIt: That is, I don't find an adequate explanation. All I find is discussions of different people's results.

First, I don't understand why, when using the FitCD methods, TMPGEnc continues to encode the entire clip instead of the 100 1-second clips or whatnot (EDIT: nevermind -- I get this now. It shows it's encoding the whole thing but in reality it's just encoding the sample). Second, I still don't completely understand how to use the Sample Size to predict the full size. Third, I don't get how you are achieving such good looking video using CQ of 25% or whatever

Can you give me a quick rundown (ie: step-by-step), starting with what to do with my AVI, how to set-up FitCD (editing FitCD.ini) to get the proper AviSynth script (including loading plugins and which plugins you use), how to encode the sample, and what to do with the sample once I've got it, and anything else I need to know? Or at least point me in the right direction. I rip quite a bit of DVD's and would LOVE to know how to fit a whole movie on one cd vs. two or three.. I've been trying this for days and I'm not having much luck.

Thanks.

Also, whenever I load the KVCDx3 template, I get a GOP of 1/5823/3/18.. upon reading http://www.kvcd.net/forum/viewtopic.php?t=1969, I thought the GOP was supposed to be 1/36/3/36..?

[SansGrip]

First, I don't understand why, when using the FitCD methods, TMPGEnc continues to encode the entire clip instead of the 100 1-second clips or whatnot

This is addressed in the thread, but it's hard to find. You're right -- the material needs to be collated and written up in an easy-to-access manner. I might have a stab at this later .

As far as your question goes, you need to use TMPGEnc's source range and hit "default" then "move to end frame" then "set end frame". For some reason it will try to encode the whole thing if you don't do this.

Third, I don't get how you are achieving such good looking video using CQ of 25% or whatever

I just got very good results with a CQ_VBR of about 9. 1h35m on one disc, almost indistinguishable from the source. It's magic .

I rip quite a bit of DVD's and would LOVE to know how to fit a whole movie on one cd vs. two or three.. I've been trying this for days and I'm not having much luck.

Keep trying. You will figure it out and it's definitely worth it...

Regards,
SansGrip

[gonzopdx]

As far as your question goes, you need to use TMPGEnc's source range and hit "default" then "move to end frame" then "set end frame". For some reason it will try to encode the whole thing if you don't do this.

Thanks, I'll give that a shot.

I just got very good results with a CQ_VBR of about 9. 1h35m on one disc, almost indistinguishable from the source. It's magic .

PLEASE explain how! Could you write up a quick/dirty tutorial a-la Doom9 FAQ's?

Keep trying. You will figure it out and it's definitely worth it...

I will.. but I'm not too hopeful. Without knowing what I'm doing I'm likely to end up wasting a ton of time with no useful results. I need something to learn from. I'm still trying to figure out how to get FitCD to load your AviSynth plugins automatically (is this necessary?) and getting the basics down.

[SansGrip]

PLEASE explain how! Could you write up a quick/dirty tutorial a-la Doom9 FAQ's?

That's something that definitely needs to be done, and really shouldn't take all that long. I'm currently testing a new GOP and file size prediction formula, but I'm thinking I could probably get a basic guide out by the end of the day. When I release it it'll be to the KVCD forum, so check there .

I'm still trying to figure out how to get FitCD to load your AviSynth plugins automatically (is this necessary?) and getting the basics down.

FitCD can't load extra plugins. I mainly only use it for its resizing capabilities (with which of course it does an excellent job), and then simply transfer the final values into my own script. I'll include all that in the guide.

[gonzopdx]

okay, so i think i figured it out.. and i'm amazed at the quality thus far =) I'm just using 0.2 of your Predictor program (which is working just dandy.. a LOT better than 0.1b =)

Wish me luck!

I will be checking the KVCD forums often from now on.. I'm especially interested in new GOP structures and etc.

[SansGrip]

okay, so i think i figured it out.. and i'm amazed at the quality thus far =)

Great . As you say, it is amazing when you consider just how high a compression ratio we're getting with these one-disc encodes.

That said, it's been a long time since the MPEG-1 standard was finalized and great advances have been made in optimized quantization matrices and GOP structures since then. These two factors (along with careful, targeted preprocessing) are absolutely the most important variables in MPEG encoding (in any DCT- and temporal difference-based compression scheme, really) as far as obtaining very high compression ratios, and the MPEG-1 standard non-intrablock matrix is especially non-optimal. I'm not really sure what they were thinking, to be honest.

So don't let anyone tell you that these templates are "just" custom matrices and GOPs -- that's the very key to high-quality low-size compression and a vast amount of theory and testing has gone into getting these kinds of results. You will not achieve even reasonably acceptable one-disc encodes of complete movies without using these matrices and GOP structures. That's what tends to confuse people when they say it can't be done.

I'm just using 0.2 of your Predictor program (which is working just dandy.. a LOT better than 0.1b =)

Glad to hear it's working for you. If my current tests turn out successful you can expect a fairly major revision for the next release.

I will be checking the KVCD forums often from now on.. I'm especially interested in new GOP structures and etc.

Definitely check it out -- some very interesting reading about the more advanced end of TMPGEnc tweaking. I also hope you and kwag can work out why you're not able to post there -- the more input we get, the harder we can squeeze that data .

Regards,
SansGrip

[gonzopdx]

If my current tests turn out successful you can expect a fairly major revision for the next release.

Sweet deal! GOP Predictions? <drool>

Out of curiosity, thus far, what are your findings for optimal GOP for the KVCDx3 templates (528x480), or for 480x480? Or what's the correlation thus far between the Quality factor and resolution? I'm halfway through the discussion thread and it seems that the lower the resolution, the tighter the GOP should be.. so if I'm doing 480x480 at around CQ_VBR12, would I be better off using the 1/15/3/1/15 or maybe something smaller? or should I stick to the default 1/5823/3/1/18 from the KVCDx3 template?

Thanks, this stuff is fascinating =)

EDIT: N/m.. read the rest of it, looks like 1/12/2/1/24 is the way to go. I'm going to try squeezing Die Another Day onto one disc @ 480x480 (since that's what my source is encoded at -- using a higher res would be pointless) and see how it turns out =)

[SansGrip]

Sweet deal! GOP Predictions? <drool>

Nope, but it should be about as accurate as I can make it. I'm also going to rework the interface into more of a wizard style.

N/m.. read the rest of it, looks like 1/12/2/1/24 is the way to go.

Yes, I'm getting very good results with that GOP.

I'm going to try squeezing Die Another Day onto one disc @ 480x480 (since that's what my source is encoded at -- using a higher res would be pointless) and see how it turns out =)

Don't forget that the quality you achieve can only be as good as the source you're using. Depending on the quality of the original encoding, the chances are you'll get much better results coming from DVD or hi-res capture than from SVCD or other low-bitrate source.

Regards,
SansGrip

[kwag]

I'm going to try squeezing Die Another Day onto one disc @ 480x480 (since that's what my source is encoded at -- using a higher res would be pointless) and see how it turns out =)

Hi gonzopdx,

You might not get good results if you're going to re-compress something which already has some layers of compression ( DVD-->DivX->SVCD ) from the original source. But hey, I do that too if it's the only source I have to convert

Edit: I should have read SansGrip's post above. It mentions what I just posted

-kwag

[gonzopdx]

Thanks for all your help guys; one last question. I tried posting this on the kvcd.net forums with no luck AGAIN. So, I've just copied/pasted what I was trying to post there. I also posted a similar message under the Editing/Cutting/Joining forum here.

- - - - -

Alright, so I'm finally figuring out this KVCD stuff =)

I'm going to try to encode a 132 minute video to once disc and see what kinds of results I can achieve @ 480x480

I have one question though that I have not been able to figure out yet.. (I also posted this on vcdhelp at this url as I've had some trouble posting here before.. (I think my ISP has routing issues).

I am trying to re-encode an SVCD which is split into three parts to one disc, but the three parts have overlapping segments. I have created a .d2v and figured the cut points perfectly. The AviSynth script I'm using is below:

LoadPlugin("C:\Program Files\Video\DVD2SVCD\MPEG2Dec\MPEG2DEC2.dll")
LoadPlugin("E:\Download\MPEG Conversion\KVCD\AviSynth\fluxsmooth.dll")
LoadPlugin("E:\Download\MPEG Conversion\KVCD\AviSynth\blockbuster.dll")
LoadPlugin("E:\Download\MPEG Conversion\KVCD\AviSynth\legalclip.dll")
# LoadVFAPIPlugin("c:\program files\video\tmpgenc\m2v.vfp","m2vsource")

mpeg2source("c:\temp\dad\dad.d2v")

# clip1 = ConvertToYUY2(m2vsource("c:\temp\dad\1\01.mpg"))
# clip2 = ConvertToYUY2(m2vsource("c:\temp\dad\1\02.mpg"))
# clip3 = ConvertToYUY2(m2vsource("c:\temp\dad\1\03.mpg"))

# Trim(clip1,343,80553) + Trim(clip2,40,80433) + Trim(clip3,14,77308)

# FlipVertical()

Trim(275,64446) + Trim(64507,128822) + Trim(128883,190450)

LegalClip()
FluxSmooth()

BB_Resolution = 480 * 480
BB_StrengthConstant = 480 * 480 * 15
StrengthValue = round(BB_StrengthConstant / BB_Resolution)

BlockBuster(method="noise", detail_min=1, detail_max=10, variance=1, cache=1024)
BlockBuster(method="sharpen", detail_min=20, detail_max=90, strength=StrengthValue)

LegalClip()

You can see in the script above my tinkering with using three seperate clips vs. the .d2v file. I've decided to use the .d2v as it removes the interlaced fields and seems to work quite a bit quicker in TMPGEnc.

My question is, how can I split the audio WITH the video? The .mpa that DVD2AVI creates contains the overlapping bits.. I want to cut those out accurately.. how can I do that?

HELP!

Thanks! And keep up the good work.. you guys are gods.

- - - - -

[SansGrip]

My question is, how can I split the audio WITH the video? The .mpa that DVD2AVI creates contains the overlapping bits.. I want to cut those out accurately.. how can I do that?

If I were you I'd make three DVD2AVI project files and corresponding WAVs, one for each MPEG-2 file. Then do something like:

part1 = AudioDub(Mpeg2Source("part1.d2v"), WavSource("part1.wav"))
part2 = AudioDub(Mpeg2Source("part2.d2v"), WavSource("part2.wav"))
part3 = AudioDub(Mpeg2Source("part3.d2v"), WavSource("part3.wav"))
part1.Trim(...) + part2.Trim(...) + part3.Trim(...)

At this point you should have the three concatenated parts without overlaps and hopefully in-sync audio.

Let me know if you can't get it working.

Regards,
SansGrip

[gonzopdx]

Actually, I figured it out before you got a chance to post 8)

What I did was:

LoadPlugin(...) 

video=mpeg2source("c:\temp\dad\dad.d2v")
audio=wavsource("c:\temp\dad\whole.wav")

AudioDub(video,audio)

Trim(video,275,64446) + Trim(video,64507,128822) + Trim(video,128883,190450) 

Plugins(...)

So, it dubbed the audio, THEN trimmed it.

I opened this script in TMPGEnc and saved the corresponding WAV, listened to it to ensure proper cutting (sounded perfect!), then removed references to the audio in the script.

I'm now playing with the samples so I can begin encoding. Whee!

[SansGrip]

Actually, I figured it out before you got a chance to post 8)

Cool! It always feels better to figure something out for yourself .

I opened this script in TMPGEnc and saved the corresponding WAV, listened to it to ensure proper cutting (sounded perfect!)

A useful piece of intelligence -- I shall try to remember that technique. Incidentally, did you check that the audio is in sync?

then removed references to the audio in the script.

For future reference, you don't need to remove the references to the audio. When you generate your video in TMPGEnc you should use "ES (Video only)" so whether audio is there or not won't matter.

If for some reason you do need to actually remove the audio, though, it would be quicker to simply use the KillAudio filter at the appropriate point in the script.

Good luck with your samples .

Regards,
SansGrip

[gonzopdx]

Incidentally, did you check that the audio is in sync?

Yep, sure did! I encoded a little 1.5 minute test right where the two files split.. perfect sync! Yippee!