digitalFAQ.com Forums [Archives] - New 10 % prediction method ?

digitalFAQ.com Forums [Archives] (http://www.digitalfaq.com/archives/)

- Avisynth Scripting (http://www.digitalfaq.com/archives/avisynth/)

- - New 10 % prediction method ? (http://www.digitalfaq.com/archives/avisynth/2841-10-prediction-method.html)

hedix

03-02-2003 06:31 PM

New 10 % prediction method ?

I think I have to put these lines at the end of my script

Code:

oldfps = framerate 

AssumeFps(oldfps * 10) 

Sampler(length=24) 

AssumeFps(oldfps)

Is this correct ?

Thank you in advance

kwag	03-02-2003 06:35 PM

That will sure work :mrgreen:

-kwag

hedix

03-02-2003 06:50 PM

Hey Kwag, you are the quickest !!!

I have done some tests, and I got this

fps x 20 , CQ=79.571 8) , 4 tries
fps x 10 , CQ=67.125, 3 tries
fps x 5 , CQ=72.020, 5 tries
fps x 1 , CQ=71.075, 3 tries

Edit: fps x 2, CQ=68.450, 4 tries

film was spanish "El 6º día", Arnold Swatzeneger, cloning ...

What's your opinion ?

kwag	03-02-2003 07:03 PM

Quote:

Originally Posted by hedix

Hey Kwag, you are the quickest !!!

I have done some tests, and I got this

fps x 20 , CQ=79.571 8) , 4 tries
fps x 10 , CQ=67.125, 3 tries
fps x 5 , CQ=72.020, 5 tries
fps x 1 , CQ=71.075, 3 tries

film was (spanish "El 6º día", Arnold Swatzeneger, clones ...)

What's your opinion ?

Run a full sampler, and when you get the wanted file size, see what CQ was needed and to what CQ is closest on one of the four CQs above :idea:

-kwag

hedix

03-02-2003 07:08 PM

Quote:

Originally Posted by kwag

Run a full sampler, and when you get the wanted file size, see what CQ was needed and to what CQ is closest on one of the four CQs above :idea:

What's a full sampler? Do you mean

Code:

Sampler(24) ?

For me is the same than "fps x 1". You know, the "fps x 10" is your 10% prediction, my "fps x5" would be a 20% prediction, and so on. :wink:

Regards

hedix

03-02-2003 07:11 PM

Ops...

On the 10 % prediction, I have to do :

Samplesize*10, or is there any other factor implied ?

kwag	03-02-2003 07:18 PM

Hi hedix,

The 10% thing is really to try to find a close CQ in many short runs. But then, a full sampler must be run. The rule is simple: The more samples taken, the more accurate the prediction. If you could start your prediction with 5% samples, and then when you get to ~3% of final size, switch to full sampler. That's the way it should work. This way, you probably run three to five 5% samplers, and then you switch to full sampler. Then it should be just one or two tries to hit final wanted file size.

-kwag

hedix

03-02-2003 07:26 PM

OK, Kwag.

I will try to implement it on ToK, and I will tell you on 3-4 days.

Thank you.

jorel

03-02-2003 08:11 PM

Quote:

Originally Posted by hedix

OK, Kwag.

I will try to implement it on ToK, and I will tell you on 3-4 days.

Thank you.

3-4 days?

i can't wait . :bawl:

:wink:

hedix

03-04-2003 07:27 AM

Hi,

The thing is working, but I'm not plenty satisfied with the result.

I think a very important thing is the initial CQ. I know It depends on the resolution, the length and the type of movie.

Could we get a formula with (aproximated) values ? Something like:

CalcResolution=Height*Width
EstimatedCQ=fx(CalcResolution,length,typeOfFilm)

Obviously, let the user choose the typeOfFilm on the program. Computers are not so smart, by now ;-)

Regards