04-26-2004, 06:38 PM
|
Free Member
|
|
Join Date: Mar 2003
Location: IamCanadian
Posts: 848
Thanks: 0
Thanked 0 Times in 0 Posts
|
|
@ BP
One more question about the guide you wrote up for me,you say min bitrate from CQ matic -do you mean use the *57 average bitrate from Calcumatic.
I used to use the *57 average with TOK and got great results but i thought i was losing quality from using such a high min bitrate
Tom
|
Someday, 12:01 PM
|
|
Site Staff / Ad Manager
|
|
Join Date: Dec 2002
Posts: 42
Thanks: ∞
Thanked 42 Times in 42 Posts
|
|
|
04-26-2004, 10:17 PM
|
Free Member
|
|
Join Date: Jul 2002
Posts: 1,224
Thanks: 0
Thanked 0 Times in 0 Posts
|
|
@bigggt,
Tom wrote:
Quote:
One more question about the guide you wrote up for me,you say min bitrate from CQ matic -do you mean use the *57 average bitrate from Calcumatic.
I used to use the *57 average with TOK and got great results but i thought i was losing quality from using such a high min bitrate
Tom
|
I use CalcuMatic with (0.57*average bitrate) = min bitrate for Tmpgenc
I also use 2000 instead of 1800 for max bitrate in Tmpgenc
This is, of course, after I load KVCD (352x240) Plus template.
Depending on the movie time, I set CQ in Tmpgenc at 50 for +120 min
and 60 or 70 for 90 minutes or less. This assumes shorter movies
will get a higher CQ and longer will get less. Full screen vs Wide screen
is considered. If a movie is widescreen say 90 minutes, I'll start CQ at 70.
If a movie is fullscreen and 90 minutes, I'll start at 60. This will
help CQTester to reach its prediction much quicker. After many encodes
you'll get a good feel about where to start CQ Again, don't forget
to do Settings --> Advanced Tab --> source range --> etc.
BTW, here's an important setting I found for CQTester. Under Settings -->
Use Dynamic test settings --> number of GOP's = 1. Use this for GOP's
that are 24 as KVCD is set to. This made my prediction much more
accurate. Also, for MB Multiplex (add multiplex size) use 12 for Mplex.
-BP
|
04-27-2004, 11:47 AM
|
Free Member
|
|
Join Date: Apr 2004
Posts: 189
Thanks: 0
Thanked 0 Times in 0 Posts
|
|
This may seem silly, but I've found a good semi-random way to predict.
Using Drp with mencoder, I set up a 2% prediction fake avi and bat file, run it, get the results, then apply the results to a 3% prediction, then apply those results to a 4% prediction.
Each prediction uses different frames for the encoding so results do vary, but an average of the three Q predicts gives quite fine results.
Does that seem silly?
__________________
|
Meeow!
|
04-27-2004, 06:01 PM
|
Free Member
|
|
Join Date: Mar 2003
Location: IamCanadian
Posts: 848
Thanks: 0
Thanked 0 Times in 0 Posts
|
|
Thanx BP
|
04-27-2004, 09:59 PM
|
Free Member
|
|
Join Date: Jul 2002
Posts: 1,224
Thanks: 0
Thanked 0 Times in 0 Posts
|
|
@Fluffbutt,
Fluffbutt wrote:
Quote:
This may seem silly, but I've found a good semi-random way to predict.
Using Drp with mencoder, I set up a 2% prediction fake avi and bat file, run it, get the results, then apply the results to a 3% prediction, then apply those results to a 4% prediction.
Each prediction uses different frames for the encoding so results do vary, but an average of the three Q predicts gives quite fine results.
Does that seem silly?
|
CQTester file prediction uses offset starting points to begin sample
frames testing at say 2% and then starting at a new frame offset for
another sample. Then it compares each for whether it comes close to
the predicted file size. If one is closer, it stays with that for the new
calculated CQ. The estimated tries are from 5 to 8 times. You can
increase the sample % but you also make the testing longer. 2% seems
to be a good sample.
-BP
|
04-28-2004, 05:27 AM
|
Free Member
|
|
Join Date: May 2003
Location: Germany
Posts: 3,189
Thanks: 0
Thanked 0 Times in 0 Posts
|
|
Quote:
Originally Posted by Fluffbutt
Using Drp with mencoder, I set up a 2% prediction fake avi and bat file, run it, get the results, then apply the results to a 3% prediction, then apply those results to a 4% prediction.
|
You dont need Drpenc for that as Avalon/Amnon just took my slicer() avsi prediction for that! He didn't asked me but he leaved the credits in my function so its ok, and thats what counts.
(Does he mention the origin in his GUI, or did he implementaed a credit button ? *lol* )
Slicer(MoviePercent, Goplenght, Offset, GoplengthMultiplikator, SubstreamOffsetcalculator)
Slicer(2,15,05,2,0)
Means:
- 2% Of the Movie will be predicted
- the GopLength in here is 15 (PAL Standard) at KVCD you would use 24
- The offset!! And here you will play! Means the offset of the slices where that whole logic bases on. In here you can differ to get various sampling turns! Useful is NOT to set an Offset of 00 as the beginning of the Movie most times is just dark and wouldnt give a good average!
- The length of each slice multiplied by the Goplentgh
- An internal calculator to get the offset value IF just doing ping pong.
So just do for example on a 2% prediction (not more is needed as you will playing with offsetted turns!!!) with various slicings:
Turns:..... Values:......................
......a)..... Slicer(2,15,05,2,0) ---> After result adjust the Encoder (Q or CQ)
......b)..... Slicer(2,15,15,2,0) ---> After result adjust the Encoder (Q or CQ)
......c)..... Slicer(2,15,35,2,0) ---> After result adjust the Encoder (Q or CQ)
......d)..... Slicer(2,15,25,2,0) ---> After result adjust the Encoder (Q or CQ)
......e)..... Slicer(2,15,05,2,0) ---> Resulted filesize ok? If not continue!
thats the approach .... and thats why you dont need more movie% on each turn! So this does minimize the whole prediction job, as every turn juses for approaching diff offsets.
Conclusion: Its better to do more offset turns then increase the predicted % of the movie! As by just differ in offsets you will effectively predict more differing parts of the movie ... in the same time of the whole job!
PS: you can also do differ in the Turns in changing the GOPmultiplicator from 2 to 1 ... means more slices but shorter.
This as seen as a whole gives you a almost randomly slicing of the movie you finally want to encode.
Thats also why just Ping-Pong (the classic way) already gives me 98% accurate encodings .... seen in a whole of all my encoded movies if jusing just Quality based setted VBR (CQ or Q).
Inc.
|
04-28-2004, 07:50 AM
|
Free Member
|
|
Join Date: Jul 2002
Posts: 1,224
Thanks: 0
Thanked 0 Times in 0 Posts
|
|
@inc,
Inc wrote:
Quote:
You dont need Drpenc for that as Avalon/Amnon just took my slicer() avsi prediction for that! He didn't asked me but he leaved the credits in my function so its ok, and thats what counts.
(Does he mention the origin in his GUI, or did he implementaed a credit button ? *lol* )
Slicer(MoviePercent, Goplenght, Offset, GoplengthMultiplikator, SubstreamOffsetcalculator)
Slicer(2,15,05,2,0)
Means:
- 2% Of the Movie will be predicted
- the GopLength in here is 15 (PAL Standard) at KVCD you would use 24
- The offset!! And here you will play! Means the offset of the slices where that whole logic bases on. In here you can differ to get various sampling turns! Useful is NOT to set an Offset of 00 as the beginning of the Movie most times is just dark and wouldnt give a good average!
- The length of each slice multiplied by the Goplentgh
- An internal calculator to get the offset value IF just doing ping pong.
So just do for example on a 2% prediction (not more is needed as you will playing with offsetted turns!!!) with various slicings:
Turns:..... Values:......................
......a)..... Slicer(2,15,05,2,0) ---> After result adjust the Encoder (Q or CQ)
......b)..... Slicer(2,15,15,2,0) ---> After result adjust the Encoder (Q or CQ)
......c)..... Slicer(2,15,35,2,0) ---> After result adjust the Encoder (Q or CQ)
......d)..... Slicer(2,15,25,2,0) ---> After result adjust the Encoder (Q or CQ)
......e)..... Slicer(2,15,05,2,0) ---> Resulted filesize ok? If not continue!
thats the approach .... and thats why you dont need more movie% on each turn! So this does minimize the whole prediction job, as every turn juses for approaching diff offsets.
Conclusion: Its better to do more offset turns then increase the predicted % of the movie! As by just differ in offsets you will effectively predict more differing parts of the movie ... in the same time of the whole job!
PS: you can also do differ in the Turns in changing the GOPmultiplicator from 2 to 1 ... means more slices but shorter.
This as seen as a whole gives you a almost randomly slicing of the movie you finally want to encode.
Thats also why just Ping-Pong (the classic way) already gives me 98% accurate encodings .... seen in a whole of all my encoded movies if jusing just Quality based setted VBR (CQ or Q).
Inc.
|
I saw your slicer prediction method in DRPenc, but didn't have a guide
to make it work until now. CQTester takes a similar approach and is
automated. (i.e. starting offset, gop setting, random samples being
compared, etc.) I don't believe anyone expects 100% accuracy, but
98% accurate is great I especially like your logic about
starting frame away from the beginning of a movie Makes good
sense.
-BP
|
04-28-2004, 09:45 AM
|
Free Member
|
|
Join Date: May 2003
Location: Germany
Posts: 3,189
Thanks: 0
Thanked 0 Times in 0 Posts
|
|
Quote:
Originally Posted by black prince
CQTester takes a similar approach and is
automated. (i.e. starting offset, gop setting, random samples being
compared, etc.)
|
YEP!
Borax and I we know each other from MVCD board, and the "dynamic" settings (gop length multi /percentual prediction based) where a suggestion by me which he integrated it
|
04-28-2004, 10:34 PM
|
Free Member
|
|
Join Date: Apr 2004
Posts: 189
Thanks: 0
Thanked 0 Times in 0 Posts
|
|
incredilbe, you live up to the name - I'd missed that bit in Drp.
Seems my idea is already in use.. LOL... (feel kinda foolish now... )
__________________
|
Meeow!
|
04-29-2004, 05:04 PM
|
Free Member
|
|
Join Date: May 2003
Location: Germany
Posts: 3,189
Thanks: 0
Thanked 0 Times in 0 Posts
|
|
Quote:
Originally Posted by Fluffbutt
incredilbe, you live up to the name - I'd missed that bit in Drp.
Seems my idea is already in use.. LOL... (feel kinda foolish now... )
|
So what keep your "engines" warm ....
|
All times are GMT -5. The time now is 07:32 PM — vBulletin © Jelsoft Enterprises Ltd
|