Quantcast New Highly Accurate Prediction Method Reported by Jorel! - digitalFAQ.com Forums [Archives]
  #1  
07-04-2003, 01:48 AM
kwag kwag is offline
Free Member
 
Join Date: Apr 2002
Location: Puerto Rico, USA
Posts: 13,537
Thanks: 0
Thanked 0 Times in 0 Posts
First of all, thanks to jorel for finding this out and doing many many encodes to confirm this.

When using ToK, or even manual KVCD prediction ( ), instead of using 24 frames for the length of the sample (Or Auto mode), use 3* frame rate.
So if you are using ToK, go to "Prediction & Spacing" and uncheck Auto, and if you are encoding at 23.976 fps, change the length of samples to 72. If you are encoding PAL (25fps), change it to 75. If encoding NTSC 29.97fps, change it to 90.
Now see how accurate your results are

Thanks jorel

-kwag
Reply With Quote
Someday, 12:01 PM
admin's Avatar
Site Staff / Ad Manager
 
Join Date: Dec 2002
Posts: 42
Thanks: ∞
Thanked 42 Times in 42 Posts
  #2  
07-04-2003, 02:19 AM
Krassi Krassi is offline
Free Member
 
Join Date: Mar 2003
Location: Germany
Posts: 390
Thanks: 0
Thanked 0 Times in 0 Posts
Quote:
Originally Posted by kwag
First of all, thanks to jorel for finding this out and doing many many encodes to confirm this.

When using ToK, or even manual KVCD prediction ( ), instead of using 24 frames for the length of the sample (Or Auto mode), use 3* frame rate.
So if you are using ToK, go to "Prediction & Spacing" and uncheck Auto, and if you are encoding at 23.976 fps, change the length of samples to 72. If you are encoding PAL (25fps), change it to 75. If encoding NTSC 29.97fps, change it to 90.
Now see how accurate your results are

Thanks jorel

-kwag
Hi kwag, hi Jorel,

that is exactly what i already tried (see here: http://www.kvcd.net/forum/viewtopic....4614&start=129)
Code:
 had to reinstall my Windows (even manual repair failed) so i've only done 2 more tests, but i had an idea:
Why not taking a longer sample size to make a better prediction with the adaptive script I had good results with a Sampler length of 50, maybe 48 would be better. I thought that a bigger sample size than GOP would screw up Prediction or Sampler() but it didn't.
Well, for me (PAL) a Sampler Length of 24 is optimal at the moment. Sometimes a SL of 72 or 75 Prediction is very accurate, sometimes way off target (mostly with 75) . Until now i made 6 test encodes with a SL of 24 and it ended always up with 0.5% accuracy.
I will try again with a SL of 72 tonight .
Reply With Quote
  #3  
07-04-2003, 02:31 AM
kwag kwag is offline
Free Member
 
Join Date: Apr 2002
Location: Puerto Rico, USA
Posts: 13,537
Thanks: 0
Thanked 0 Times in 0 Posts
For PAL (25fps) try 75, which is exactly 3*frame rate.

-kwag
Reply With Quote
  #4  
07-04-2003, 10:20 AM
girv girv is offline
Free Member
 
Join Date: Sep 2002
Posts: 108
Thanks: 0
Thanked 0 Times in 0 Posts
Send a message via ICQ to girv
I tried this on a couple of encodes and still got 2-3% error. Can you post a screenshot of the ToK settings to use? Ta.
Reply With Quote
  #5  
07-04-2003, 08:44 PM
labomba labomba is offline
Free Member
 
Join Date: Jan 2003
Location: S.Paulo-Brasil
Posts: 185
Thanks: 0
Thanked 0 Times in 0 Posts
Kwag and Jorel, please help me.

Look this TOK log from "Heaven":

================================================== ===========
ToK Log: E:\HEAVEN\VIDEO_TS\NewScript25.avs
================================================== ===========

Resolution (fps):704x480 (23,976 fps)
Total Frames: 139897
Total Time : 01:37:15
-------------------------------------------------------------

Audio Size: 93.359.125
Required Video Size: 733.099.284

Factor: 59,940
Desired Sample Size: 12.230.551

-------------------------------------------------------------
New Faster Prediction
-------------------------------------------------------------

Full Sample
Next CQ: 70,000. Sample Size: 16.014.598
Small Sample
Next CQ: 70,000. Sample Size: 1.581.155
Predicting...
Next CQ: 53,460. Sample Size: 12.035.700
Next CQ: 58,000. Sample Size: 13.587.551
Next CQ: 54,880. Sample Size: 12.048.249
Next CQ: 55,845. Sample Size: 12.056.600
Next CQ: 56,506. Sample Size: 12.717.480
Next CQ: 56,097. Sample Size: 12.274.154

Exit Condition: 0,500 % reached ! yahoo !
Tries : 7


Final CQ: 56,097
Total Time For Predicition: 00:17:48


-------------------------------------------------------------
Encoding E:\HEAVEN\VIDEO_TS\NewScript25.avs
-------------------------------------------------------------

Encoding... CQ : 56,097
Final Encoded Size: 684.997.229
Muxing ...


Total Time (all operations): 04:47:37

Finished

I don't understand
Required Video Size: 733.099.284
Final Encoded Size: 684.997.229

My settings:
TOK 0.0.5.2 with Kwags video.en1 modified. Also, I changed the VBV Buffer Size to 60960 instead 0f 40960 (illegal floating zero trick).

Prediction&Spacing:
New Faster Prediction : checked
Factor for Prediction : 1,01
Length of Sample : 24 (Auto checked !)
Number of Samples / minute : 2
1st Group Precision % : 0,50
Speed Up by: 10
MPEG1 - VCD

Any idea?

LaBomba
Reply With Quote
  #6  
07-04-2003, 09:18 PM
jorel jorel is offline
Invalid Email / Banned / Spammer
 
Join Date: Aug 2002
Location: Brasil - MG - third stone from the sun
Posts: 5,570
Thanks: 0
Thanked 0 Times in 0 Posts
labomba my friend:


change the
Length of Sample : 24 (Auto checked !) to 72 (uncheck Auto !)

Number of Samples / minute : 1
Factor for Prediction : 1,00

see the first Kwag post!


for vbv buffer size i'm in doubt ...
seems too much!
wait for the Kwag recomendations!

Reply With Quote
  #7  
07-04-2003, 09:18 PM
kwag kwag is offline
Free Member
 
Join Date: Apr 2002
Location: Puerto Rico, USA
Posts: 13,537
Thanks: 0
Thanked 0 Times in 0 Posts
Quote:
Originally Posted by labomba

Prediction&Spacing:
New Faster Prediction : checked
Factor for Prediction : 1,01
Length of Sample : 24 (Auto checked !)
Number of Samples / minute : 2
1st Group Precision % : 0,50
Speed Up by: 10
MPEG1 - VCD

Any idea?

LaBomba
Please try the settings that we mention in this thread. You have a Factor of 1.01 instead of 1.0.
You have to uncheck Auto, and set your Length of sample to 3*FrameRate.
Set number of samples per minute to 1.

Now try it again

-kwag
Reply With Quote
  #8  
07-04-2003, 09:30 PM
ovg64 ovg64 is offline
Free Member
 
Join Date: Jan 2003
Location: Puerto Rico
Posts: 423
Thanks: 0
Thanked 0 Times in 0 Posts
Send a message via MSN to ovg64
Would that make up on that 20-30mb or 0.30CQ difference
Reply With Quote
  #9  
07-04-2003, 09:59 PM
labomba labomba is offline
Free Member
 
Join Date: Jan 2003
Location: S.Paulo-Brasil
Posts: 185
Thanks: 0
Thanked 0 Times in 0 Posts
I made a new try but BEFORE to read these last posts, so:

Factor REMAINS 1,01 and Samples/minute = 2

================================================== ===========
ToK Log: E:\HEAVEN\VIDEO_TS\NewScript25.avs
================================================== ===========

Resolution (fps):704x480 (23,976 fps)
Total Frames: 139897
Total Time : 01:37:15
-------------------------------------------------------------

Audio Size: 93.359.125
Required Video Size: 733.099.284

Factor: 19,980
Desired Sample Size: 36.691.657

-------------------------------------------------------------
New Faster Prediction
-------------------------------------------------------------

Full Sample
Next CQ: 70,000. Sample Size: 48.104.229
Small Sample
Next CQ: 70,000. Sample Size: 4.761.502
Predicting...
Next CQ: 53,393. Sample Size: 36.235.344
Next CQ: 57,864. Sample Size: 40.780.512
Next CQ: 54,735. Sample Size: 36.268.157
Next CQ: 55,664. Sample Size: 36.291.980
Next CQ: 56,312. Sample Size: 37.658.931
Next CQ: 55,921. Sample Size: 36.320.490
Next CQ: 56,073. Sample Size: 36.840.266

Exit Condition: 0,500 % reached ! yahoo !
Tries : 8


Final CQ: 56,073
Total Time For Predicition: 00:49:44

Total Time (all operations): 00:49:44

Finished

Look: FINAL CQ: 56,073 , lower than the first try (56,097)

Do you think that changing the Factor to 1.0 and the samples/minute to 2, will increase the Final CQ?

LaBomba
Reply With Quote
  #10  
07-04-2003, 10:05 PM
kwag kwag is offline
Free Member
 
Join Date: Apr 2002
Location: Puerto Rico, USA
Posts: 13,537
Thanks: 0
Thanked 0 Times in 0 Posts
Hi LaBomba,

I don't know what's going on with your source
You're not encoding an AVI are you

-kwag
Reply With Quote
  #11  
07-04-2003, 10:13 PM
labomba labomba is offline
Free Member
 
Join Date: Jan 2003
Location: S.Paulo-Brasil
Posts: 185
Thanks: 0
Thanked 0 Times in 0 Posts
No Kwag.

I only use DVD material.
I used ForceFilm in DVD2AVI (Film 98% Progressive).
Used the optimal script from July 2th to AVS 2.52
My TMPGEnc is 2.513

Please Help

LaBomba

[/IMG]
Reply With Quote
  #12  
07-04-2003, 10:21 PM
kwag kwag is offline
Free Member
 
Join Date: Apr 2002
Location: Puerto Rico, USA
Posts: 13,537
Thanks: 0
Thanked 0 Times in 0 Posts
What movie are you encoding

Edit: I see it's Heaven. What type of movie is that

-kwag
Reply With Quote
  #13  
07-04-2003, 10:22 PM
jorel jorel is offline
Invalid Email / Banned / Spammer
 
Join Date: Aug 2002
Location: Brasil - MG - third stone from the sun
Posts: 5,570
Thanks: 0
Thanked 0 Times in 0 Posts
Set number of samples per minute to 1 labomba.
search the prediction CQ again.

in my system do a big difference!
Reply With Quote
  #14  
07-04-2003, 11:57 PM
ovg64 ovg64 is offline
Free Member
 
Join Date: Jan 2003
Location: Puerto Rico
Posts: 423
Thanks: 0
Thanked 0 Times in 0 Posts
Send a message via MSN to ovg64
Quote:
Originally Posted by labomba
No Kwag.

I only use DVD material.
I used ForceFilm in DVD2AVI (Film 98% Progressive).
Used the optimal script from July 2th to AVS 2.52
My TMPGEnc is 2.513

Please Help

LaBomba

[/IMG]
Maybe you should start thinking about loweing you resolution from 704 to 528
Reply With Quote
  #15  
07-05-2003, 10:57 AM
labomba labomba is offline
Free Member
 
Join Date: Jan 2003
Location: S.Paulo-Brasil
Posts: 185
Thanks: 0
Thanked 0 Times in 0 Posts
Oh guys, sorry for the delay!
Here the last try, with all the changes (Samples/min = 1, Factor=1)
Bingo
Audio Size: 93.359.125
Required Video Size: 725.840.875

Factor: 19,980
Desired Sample Size: 36.328.373

-------------------------------------------------------------
New Faster Prediction
-------------------------------------------------------------

Full Sample
Next CQ: 70,000. Sample Size: 46.664.520
Small Sample
Next CQ: 70,000. Sample Size: 4.673.724
Predicting...
Next CQ: 54,495. Sample Size: 35.417.417
Next CQ: 58,999. Sample Size: 39.851.997
Next CQ: 56,084. Sample Size: 35.983.535
Next CQ: 56,943. Sample Size: 38.331.786
Next CQ: 56,361. Sample Size: 36.823.666
Next CQ: 56,210. Sample Size: 36.369.733

Exit Condition: 0,500 % reached ! yahoo !
Tries : 7


Final CQ: 56,210
Total Time For Predicition: 23:31:36

Total Time (all operations): 23:31:36

Finished

The CQ increased from 56,073 to 56,210

A little bit low CQ yet, but I'm trying a 97min. movie, some dark but also with low action, with 704x480 resolution, 1 CD80min, 128k audio!!!
Madness ??! Really NO !! I can testify, with the second last script of July 2th, and the old CQ (56,073), I had very little gibbs.

Now, let me try the newest script and the newest prediction method.

Thanks a lot, guys.

LaBomba
Reply With Quote
  #16  
07-05-2003, 11:28 AM
ovg64 ovg64 is offline
Free Member
 
Join Date: Jan 2003
Location: Puerto Rico
Posts: 423
Thanks: 0
Thanked 0 Times in 0 Posts
Send a message via MSN to ovg64
Well im not sure this method more accurate than the other but correct me if im run chainge the sampler rate fron 24 to 72 for NTSC film here is my tok lod:

Quote:
================================================== ===========
ToK Log:
================================================== ===========

Resolution (fps):528x480 (23.976 fps)
Total Frames: 136900
Total Time : 01:35:10
-------------------------------------------------------------

Audio Size: 79,940,000
Required Video Size: 739,260,000

Factor: 19.980
Desired Sample Size: 37,000,001

-------------------------------------------------------------
New Faster Prediction
-------------------------------------------------------------

Full Sample
Next CQ: 70.000. Sample Size: 29,784,247
Small Sample
Next CQ: 70.000. Sample Size: 3,348,844
Predicting...
Next CQ: 86.959. Sample Size: 50,008,606
Next CQ: 77.265. Sample Size: 37,045,593

Exit Condition: 0.500 % reached ! yahoo !
Tries : 3
when over the desire video size by ~8 megs 728mb video, 79mb audio from HeadAC3he. Good ting i was able to Overburn, any sugestion maybe usin 64 samples instead of 72
Reply With Quote
  #17  
07-05-2003, 06:24 PM
ovg64 ovg64 is offline
Free Member
 
Join Date: Jan 2003
Location: Puerto Rico
Posts: 423
Thanks: 0
Thanked 0 Times in 0 Posts
Send a message via MSN to ovg64
Well im not sure this method more accurate than the other but correct me if im run chainge the sampler rate fron 24 to 72 for NTSC film here is my tok log:

Quote:
================================================== ===========
ToK Log:
================================================== ===========

Resolution (fps):528x480 (23.976 fps)
Total Frames: 136900
Total Time : 01:35:10
-------------------------------------------------------------

Audio Size: 79,940,000
Required Video Size: 739,260,000

Factor: 19.980
Desired Sample Size: 37,000,001

-------------------------------------------------------------
New Faster Prediction
-------------------------------------------------------------

Full Sample
Next CQ: 70.000. Sample Size: 29,784,247
Small Sample
Next CQ: 70.000. Sample Size: 3,348,844
Predicting...
Next CQ: 86.959. Sample Size: 50,008,606
Next CQ: 77.265. Sample Size: 37,045,593

Exit Condition: 0.500 % reached ! yahoo !
Tries : 3
when over the desire video size by ~8 megs 728mb video, 79mb audio from HeadAC3he. Good thing i was able to Overburn , any sugestion maybe usin 64 samples instead of 72
Reply With Quote
  #18  
07-05-2003, 07:45 PM
J-Wo J-Wo is offline
Free Member
 
Join Date: Nov 2002
Location: Toronto, Canada
Posts: 454
Thanks: 0
Thanked 0 Times in 0 Posts
Question: does this method work for fast prediction or full prediction only? Also, is it accurate at other factors for prediction? I ask because I like to make my muxed videos to be ~815mb since I can overburn on my discs. Lets me squeeze a bit more CQ in my videos, you know? So for that I usually set my factor a bit higher, to 1.01 or 1.02.

Also, is it recommended to change # of samples/min to 2 if the total movie length is less than 100min? I have heard that prediction accuracy is best when you can make at least 100 samples per movie, so if my film is only 90 minutes long I predict with 2 samples/min.
Reply With Quote
  #19  
07-05-2003, 08:46 PM
kwag kwag is offline
Free Member
 
Join Date: Apr 2002
Location: Puerto Rico, USA
Posts: 13,537
Thanks: 0
Thanked 0 Times in 0 Posts
Quote:
Originally Posted by J-Wo
is it recommended to change # of samples/min to 2 if the total movie length is less than 100min?
Yes it is
Quote:
I have heard that prediction accuracy is best when you can make at least 100 samples per movie, so if my film is only 90 minutes long I predict with 2 samples/min.
That's right

-kwag
Reply With Quote
  #20  
07-06-2003, 04:08 AM
ozjeff99 ozjeff99 is offline
Free Member
 
Join Date: May 2002
Location: Sydney, Australia
Posts: 159
Thanks: 0
Thanked 0 Times in 0 Posts
Kwag / Jorel, thanks for making things even better for us.

Could you confirm if these settings OK then for PAL?

New faster Prediciton: Unchecked
Check Pass: Unchecked
Precision: 0.5%
Speed up by: 10 times
Factor for Prediction: 1.00
Length of Sample: 75
Number of Samples: 1 For > 100mins
2nd Group of Passes: Unchecked

GOP parameters: 1, 5823, 2, 1, 24

Much appreciated.
ozjeff99
Reply With Quote
Reply




Similar Threads
Thread Thread Starter Forum Replies Last Post
CQ prediction method WOWIEGURL Avisynth Scripting 13 03-12-2005 04:56 PM
File size prediction sometimes not accurate using vmesquita? rs008f Video Encoding and Conversion 4 09-07-2003 05:09 AM
Faster prediction method ARnet_tenRA Avisynth Scripting 19 04-12-2003 09:11 AM
New 10 % prediction method ? hedix Avisynth Scripting 9 03-04-2003 07:27 AM
File Size Prediction Dosen't Seem To Be Accurate Paul0889 Avisynth Scripting 2 12-31-2002 03:50 PM

Thread Tools



 
All times are GMT -5. The time now is 01:08 AM  —  vBulletin Jelsoft Enterprises Ltd