CQMatic 1.4.00 BETA
 

Latest is RC-1, at the end of this thread.


Please test this, so I can keep tunning it for maximum accuracy.

Edit:

BETA 3:
1)Added "seconds" field. Needed when doing small videos. We needed that a LONG time ago ;)
2) Slight inteface reorder.

http://www.kvcd.net/downloads/CQMatic-1.4.00-BETA-3.exe

-Karl

BETA 4
 

Same as BETA 3, but now with user selectable samples per phase.
Automatically selects the correct (approximate) samples, depending on the movie time. The scroll bar automatically adjusts depending on what you enter.

http://www.kvcd.net/downloads/CQMatic-1.4.00-BETA-4.exe

-Karl

BETA 5
 

1) Clamped minimum samples to 10. Lower values give inaccurate results.
2) Small cosmetic fixes. Mins instead of Min, and Secs instead of Sec.

http://www.kvcd.net/downloadsCQMatic-1.4.00-BETA-5.exe

I just converted a small PAL Interlaced VOB, 4 minutes 41 seconds song "duran duran - ordinary world". (Thanks Ademar :mrgreen: )

Wanted file size: 44,456KB
Encoded file size: 44,721KB

I'm doing more heavy tests now.


Edit: I wanted to keep this private, because if I post this public, it will be a mess :roll:
So until you and I are reasonably happy with results, then I'll post it on the forum.

-Karl

Hi,

Link to beta5 don't work...

Salu2
Fabrice

Stick around :)
RC-1 coming up after more final tests and calibrations I'm doing ;)

-Karl

A taste of precision, at least on my end!
 

CQMatic 1.4.00 Release Candidate 1
(Link below, so you all have to read this :lol: )

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

PLEASE, let me know if you find any problems, cause my hands are EAGER to post this on the forum.



First Video test:
"inxs - devil inside" (Thanks again Ademar :) )
6 minutes, 30 seconds, interlaced, 29.97fps.
Target: 704x480

First, I encoded the video at a fixed CQ of 65, and noted file size. 80,145KB

Then, I ran CQMatic 4 times, for validation, and here are the results:

Run 1) CQ found: 66.75. file size: 80,460KB :arrow: 0.3914% accurate
Run 2) CQ found: 66.899: File size: 80,469KB :arrow: 0.4026% accurate
Run 3) CQ found: 67.729: File size: 80,521KB :arrow: 0.4669% accurate
Run 4) CQ found: 66.59: File size: 80,450KB :arrow: 0.3791% accurate



Secod Video test:
"duran duran - ordinary world"
4 minutes, 41 seconds, PAL Interlaced.
Target: 704x576

Run #1
Wanted target: 43,332KB

Found CQ of 5.1 8O (Don't ask, but that's what it found), and final file size: 42,257KB :mrgreen:
Guess I asked for too low file size fror clip. ( Still, :arrow: 2.5439% accuracy )

Run #2
Increased wanted target, to get higher CQ.
Wanted target: 74052
Found CQ of 89.51: Final size: 76,034KB for a :arrow: 2.6067% accuracy.



And the BIG test :!:
"Red Planet", full movie prediction, 23.976fps Progressive.
Target: 704x480
Wanted video size: 722,734KB
CQ found: 68.9 :!: :!: :!:

Encoded file size, don't care :lol:
Because with an older BETA, I got 709,325KB at a CQ=65.81 :D


Here's a copy of the log:

What's new on this version :?:
1) Completely new prediction engine: Calibration process, to determine source to sampling compressibility.
So now you should be able to throw anything at CQMatic, and it will know "how" to encode "whatever" material :D
2) Prediction is now done in "phases". Then, using a formula and some tricks (don't ask ;) ), it determines the correct target CQ.
3) Prediction is MUCH faster :!:
On my Pentium 4 @2.8Ghz, I am averaging a full prediction cycle in about 20 to 25 minutes minutes (on linear range, CQ=~50 to CQ=~85)
Below CQ=~50, well, you all know it's longer :roll:
The first phase, in theory, should be the longest, as all subsequent phases uses the previous found CQ,, which is closer to final value, instead of starting every phase at CQ=60.
4) Removed "Override CQ" option. It's now done automagically ;)
5) Added progress bar, active after calibration phase.



:!: GRAB IT :!:

http://www.kvcd.net/downloads/CQMatic-1.4.00-RC-1.exe

And please, REPORT :!:

Note: I'm really looking for consistency in this version.
That means that it should produce a very small file size deviation, no matter how many times you encode the same source.
If this does indeed succeed, then my next phase is minor tweaks to adjust offset for target accuracy.

Thanks :)
-Karl

This is one reason, why it is nice to be a mod here. It's just a pity that there's nothing to do for me in my forum :cry:

Well why do you say that 8O

-Karl

Just that german forum activity is near flatline.

Yes...

Did not mean the entire forum.

Karl, with RC-1 CQMatic took 68 minutes to find a CQ of 89.50 on my 146 movie.

The previous one took 4 hours to find (a too low) 80.6

Perhaps this time I will have something in the target :-)

Great!
Let me know the final result :)
I'm still not 100% confident of my 2%, worse case accuracy, though :roll:
So I'm taking a little longer to do more validation tests.

-Karl

Released!
Thread now continues on forum

-Karl

I still want to add my result :
Wanted size : 2 114 300
Encoded size : 2 223 532

Not sot bad :-)

Hi Phil,

Try the released version, at least prediction only, and check if it closely matches the CQ found for the one you just did.
I just had a bad experience with a "Red Planet" run, where after about 10 different runs and getting a very close CQ ( 65 to 68 ), I got one CQ of 70.2 ( :x ) which that tiddy little change boosted my final size to ~780,000KB :x
I hate that CQ range in the range of 68 to 70 :!:
It's such a PAIN, because the difference in file size is gigantic, and it's very hard to predict, unless more runs are done.
I'll probably add an option to "double" the number of phases, at the expense of longer time to predict.
But this will guarantee a higher accuracy.

-Karl