kwag:

I think you're dealing with a placebo formula :(

size/time=bitrate
size/bitrate=length

so,

MovieStacker's Elementary Video Stream / Total Minutes of Movie = MovieStacker's Average Bitrate (in kilobytes/minute)


MovieStacker's Average Bitrate (in kilobytes/minute) / MovieStacker's Average Bitrate (in kilobits/second) = 60/8.

Demuxed .m2v file size (in kilobytes)/ Average Bitrate of demuxed.m2v (in kilobits/second) = total seconds of Demuxed .m2v/8 =60/8

As total seconds of Demuxed .m2v is 60s (since you chose a 1 min sample)

CompressionFactor = 60/8 / 60/8 = 1.

Any deviance from 1.0000 is because:

1) The length of the movie in minutes is truncated (you discard the final seconds)
2) The length of the sample ain't exactly 60 seconds.

And not because of the compressability of the movie.


Sorry.

Kwag / GFR,

May be this has been done, suggested and tried before but would another way of getting the compressability of the movie be possible if we did this:

Take a sample with TMPeg with no filters or MA (just resize) using a base CQ of 70.

Then doing the sample again with TMPeg and CQ 70 using the final script (with the filters + MA). The compressability could be defined as

Sample1 Size (only resize) / Sample2 Size (Full MA script with resize)

We then use this factor in Tok to get the correct CQ? I know the CQ factor is not linear and it sounds too simplistic (not for my simple brain though :!: )

I hear what you say GFR and it makes sense, though you and kwag seem far more in tune with this stuff. I'm just bouncing some ideas so please feel free to laugh at my kindergarten suggestion. :wink:

Ignore me... :imstupid:

My "theory" has no basis. My test results on a 2.5 hour NTSC rip:

Guess I forgot the non linear behaviour of CQ.

Hmm, I have a feeling :D we should orient to the average bitrate. If the average bitrate of encoded sample = average bitrate reported for full video file that is the CQ value.

I'm heavily working on this issue ... :cool:

-kwag

Hi to all,

After doing several tests on file prediction for the past 4 days, I believe I found the problems related to the innacurate prediction.
It's all related to the dynamic range of the encoder, when using a wide range as we normally do ( 300 MIN, 2,500 MAX )
Right now, with KVCDx3 528x480(576), there's really no need to go beyond 2,000Kbps, when using the MA script.
So I have found that using a MAX of 2,000Kbps and a MIN of (averageBitrate - 43%), produces extremely accurate results.
Here are the numbers on 4 different movies:



So you see, the worst case is about ~16MB, which is really negligible in visual quality.

The method is really simple:
Set your MAX bitrate to 2,000Kbps.
Set your MIN bitrate to Average bitrate (data from Moviestacker) - 43%.
Example: My K-Pax movie average bitrate is 800Kbps. So 800 - 43% = 456Kbps
That's it :!:
Now the encoder's dynamic range will be 2000 - 456 = 1,544Kbps, instead of 2,500 - 300 = 2,200Kbps, and the results are far more accurate.
This is probably the reason why the MainConcept encoder is very poor when doing one pass encoding on resolutions greater than 352x480(576), because it's dynamic range is very narow ( way smaller than TMPEG ), but at 352x240(288) it works extremely well with a bitrate range of ~300 to ~1,500.
The file prediction data above was done manually with "Sampler()" at the end of the script. I didn't use ToK, because I wanted as consistent results as I could, so I did them with the manual method. ToK should behave the same with the changes too. Maybe now the "Fast" prediction will be as accurate as the "Full" prediction :idea:
So that has to be tested.
So there you have it :) I hope this works as good for PAL people, as it works for us here in NTSC land :)

----------------------------
@ovg64,
Thanks for your tests ;)
----------------------------

-kwag

And here are the results of the same 4 movies, but using ToK to calculate CQ in "Fast" mode with the new method described above:

All calculations were done using: one sample per minute, 24 frames per GOP (Length of sample (Auto)), factor of 1.0 and speedup of 10 (Tenra's "Fast" method)

-kwag

Just wondering... I've never looked for average bitrate in MStacker before, but I just checked out a couple of my existing d2v files and most are around 1800-2200 kbps. So why is yours for K-Pax only 800??? Am I looking in the right place?

My K-Pax movie is ~120 minutes. When I load the .d2v in MovieStacker, the suggested average bitrate is exactly 800Kbps. That's for a target of one 80 minute CD-R.
You probably have MovieStacker set to 2 CD's ;)You leave the MIN set exactly as you did with prediction. That is, -43% of average bitrate. And leave the MAX at 2,000.

-kwag

Hi jwo 800 is the average now take 800 - 43% = 456Kbps and that is the min. you are going to use, 2000 is the max. and you ll encode with those numbers too.

@Kwag,
Your prediction results look pretty good.

I`m wondering if we can get away with that low of a max bitrate,
on Mpeg2 vids... :?: (visual quality wise)

MMMMM...more tests...more tests... :)



*******************************
The Devil`s always.....in the Details!

With the MA script, yes :)

-kwag

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


*******************************
The Devil`s always.....in the Details!

It's beer time :mrgreen:
http://www.kvcd.net/boon-action.mpg

Edit: The sample is 1,115Kbps average bitrate :mrgreen:

-kwag

means:
downstairs to the kitchen again!
:lol:

I'm on my 4'th trip :mrgreen:

Wow! You guys really are amazing! :D
Could someone please post a screenshot of where to find the average bitrate in MovieStacker? And how to set up MovieStacker correctly?
I've never really used this software before... :oops:
Thanks!

audioslave,
:)

without screenshot, it's easy!

run MovieStacker and if you are using avisynth2.5x
just click "ignore" in the warning advice.

:arrow: in the "Resize & Video Stream" tab,
in "Source" under "MPEG Resizing" frame,
load your project file(*.d2v or *.avi).

:arrow: the average bitrate(kbps) is under
"Video Stream" frame in the bottom of this tab!

:wink:

Of course! Thank you very much for you help jorel! :D
Easy and simple. Like I said, I haven't been using MovieStacker that much before... :wink:

HI,does anyone know if this works with avi's because when i load them into moviestacker the average bitrate is like 1000-1300 which in turn would make the min bitrate around 600 or so.Is 600 min going to make the encode like like crap.

Thanx

It should work the same with AVIs. Same principles apply.

-kwag

It seems not to work with PAL system. :cry:
I try prediction with "The Matrix" for one CD at 704x576 (I have recently encoded it at that resolution with CQ 56):
- average bitrate 755 (cutting end credits)
- set min=430 / max 2000 (using ToK 0.0.5.2, Avisynth 2.5x and last script)
- factor for prediction 1,02 (overburning)
- new faster prediction (Tenra) with precision 0,5%
- lenght of sample 24 / nr. of sample 1
- audio 128 and GOP 25

Ended prediction with factor of apprx. 34... :oops:

Any idea? 8O

I rest with my method, working for me (PAL): standard setting in ToK (v0.0.5.2), but with nr. of sample set to "2", GOP 25 and lenght of sample 24 (works also well with fast prediction). :wink:

Yes. Do a manual prediction with "Sampler()" ;)

-kwag

kwag wrote:
With manual prediction:
frames 186314
time 02:04:13
PAL 25 frames
186314/124/25 = 60,10
800 MB - 115 MB (audio) = 685 MB (for video)
685/60,10 = 11,39 MB (size for sample)

found CQ 57,3 (at 11,4 MB) :P

Well, but now the most important thing: encoding! :D

Thanks kwag, as soon as possible I post my result... :!:

EDIT:
Well, encoding finished, final file size is 686 MB! 8O

-break

sampler prediction "script" samples..
 

Hello peoples :)

Does anyone have any good .AVS samples of the "sampler" for the
manual prediction method ??

All I have (after a careful search) is what I got (below) but I know there
are some missing pieces before and after hte Sampler() ...
Note, I'm not using MStacker (nor do I have it installed) I'm simply
trying to understand the "prediction" process :)

Thank you all,
-vhelp

Re: sampler prediction "script" samples..
 

The only thing I do, is to add as last line to my AVS script the command "Sampler(lenght=24).
Then I launch TMPGEnc to create sample video, after I found exact size I want, comment last line (sampler command) with "#", and encode the full movie.
I remember you the great Kwag guide for understand manual prediction
here: http://www.kvcd.net/forum/viewtopic.php?t=2102

I hope this can help you. :wink:

-break

hi break..

Thanks for your response. Actaully, I have read that link - many times, and
am still a bit dazed at it. But, in any case, I don't think it is used any longer,
per Kwag's recommend (elsewheres) where he (in so many words)
said, this is the new prefered method or something like that. I can't remember.

But, I read that that link has some issues w/ which page has the actual
how-to preduct process.

Sooner or later, I'm bound to get it. At the moment, I'm resarching one
too many things :)

Thanks again pal,
-vhelp