digitalFAQ.com Forums [Archives] - A new way to calculate CQ... maybe?

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

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

- - A new way to calculate CQ... maybe? (http://www.digitalfaq.com/archives/avisynth/2340-calculate-cq.html)

Of course... the audio... :x
i am usualy not using AviSource(movie, FALSE).
Anyway thanks a lot :D

Quote:

Originally Posted by kwag

I just wish CQ would be a linear scale icon_cry.gif
Then we wouldn't have all this trouble

Yeah, but what fun would that be? :mrgreen:

Ok I finally finished my spreadsheet- lot of work for file prediction :!:

I encode samples of Heist with this script:

Code:

Mpeg2Source("D:\HEIST\VIDEO_TS\heist.d2v")

LegalClip()

BilinearResize(512,344,3,0,714,480)

AddBorders(8,68,8,68)

LegalClip()

which is very basic- no noise filters

here's the result:
http://www.digitalfaq.com/archives/error.gif
here's the excel file for the Sigmoidal Curve parameter sticky post 8O: Heist_data.xls

I still think it may be possible to create a linear equation based on cq0 to ~cq50 and combine another one for cq50 to ~cq80 8O

ren

Quote:

Originally Posted by ARnet_tenRA

The problem I see with these values is that your CQ100 seem way out of proportion with the CQ0 and CQ50 values.

Using the scale I had posted before you should only have about 17.778MB and 1194 as the avg. bitrate for CQ100. ie: double CQ50

Your CQ100 is triple CQ50 and that is why the formula is not working.

Yep, that is very weird 8O I wonder if this might have to do with the resolution :?: I used the KVCDx3 528x480 template
http://www.digitalfaq.com/archives/error.gif

Quote:

Originally Posted by rendalunit

I wonder if this might have to do with the resolution :?: I used the KVCDx3 528x480 template

That is a very good possibility. I used 352x240 for my tests.

-Tenra

If this means that the curves are different for each resolution, we're screwed 8O

Quote:

Originally Posted by kwag

If this means that the curves are different for each resolution, we're screwed 8O

Don't worry, I'm working on a new formula where you take 3 or 4 short samples and the curve for your movie is automatically determined. :wink:

-Tenra

I encoded a bunch of samples of"the Matrix" at 528x480 and graphed the sizes and it follows exactly the same pattern!
http://www.digitalfaq.com/archives/error.gif
The peak file sizes (or the "ceiling") is higher with the Matrix samples 8O The bottom line is that it may be possible to determine the final size of the encode with one short sample at cq-100 and with a good formula (maybe range adjusted by resolution or number of pixels), making the process simpler :D :tongue2:

ren

OK, I'm finally done, and I think I have good news :!:

First, a chart, and then the explanation :lol: :

Code:

CQ        FS          ABR         !ABR           *CQ



100  4349        1603        1603           100

99        4336        1598        1586.97        98.32

98        4323        1593        1570.94        96.64

97        4310        1589        1554.91        94.92

96        4298        1584        1538.88        93.27

95        4287        1580        1522.85        91.56

94        4276        1576        1506.82        89.87

93        4265        1572        1490.79        88.2

92        4255        1568        1474.76        86.53

91        4246        1565        1458.73        84.82

90        4237        1561        1442.7         83.18

89        4228        1558        1426.67        81.5

88        4219        1555        1410.64        79.83

87        4211        1552        1394.61        78.18

86        4204        1549        1378.58        76.54

85        4196        1546        1362.55        74.91

84        4171        1537        1346.52        73.59

83        4032        1486        1330.49        74.31

82        3884        1431        1314.46        75.32

81        3692        1361        1298.43        77.28

80        3419        1259        1282.4         81.49

79        3085        1136        1266.37        88.07

78        2990        1101        1250.34        88.58

77        2985        1099        1234.31        86.48

76        2980        1097        1218.28        84.4

75        2975        1095        1202.25        82.35

74        2966        1092        1186.22        80.38

73        2880        1060        1170.19        80.59

72        2784        1025        1154.16        81.07

71        2613        961         1138.13        84.09

70        2402        884         1122.1         88.85

69        2362        869         1106.07        87.82

68        2361        869         1090.04        85.3

67        2356        867         1074.01        83

66        2352        866         1057.98        80.63

65        2298        845         1041.95        80.15

64        2222        817         1025.92   80.37

63        2078        765         1009.89        83.17

62        1987        731         993.86         84.29

61        1985        731         977.83         81.6

60        1983        730         961.8          79.05

59        1978        728         945.77         76.65

58        1947        717         929.74         75.21

57        1882        693         913.71         75.15

56        1773        653         897.68         76.98

55        1768        651         881.65         74.49

54        1759        647         865.62         72.25

53        1756        647         849.59         69.6

52        1742        641         833.56         67.62

51        1695        624         817.53         66.82

50        1623        597         801.5          67.13

49        1615        595         785.47         64.69

48        1613        594         769.44         62.18

47        1616        595         753.41         59.51

46        1602        590         737.38         57.49

45        1546        569         721.35         57.05

Please ignore the three columns in the center. They were used only for reference while creating the factors. The important numbers are the CQ on the left, and the *CQ on the right ONLY.

FS=File Size.
ABR=Average Bit Rate.
!AVBR=Linear(should be) Average Bit Rate.
*CQ=Use this CQ to encode.

This was done with 352x480 template to pick a "half way" point for resolution.

Now you only need to run a sampler ( ~30% or full ) one time only with CQ=100 :!:

And do this:
(1)Lookup the average bit rate that MovieStacker suggested.
(2)Divide Moviestacker's suggested average bit rate by your real average bit rate of your sample as seen on BitRate Viewer, and multiply by 100.
(3)Look up the closest number you got in step (2) IN THE LEFT CQ values column of the chart, and use the *CQ on the same line on the far right

Tada!. That's it.

Please let me know if this is on target, or if it has been a waste of time :cry:

Edit: Suggestion, I think it's better to run the one time full sampler instead of ~30%. This way we get a more accurate bit rate for the prediction. Also these numbers are to be used with KVCD's Q Matrix and GOP. They will be WAY off with anything else :!:

-kwag

@Ren,

Take a look at the numbers and see if it now matches close to what you had encoded before :wink:

-kwag

Hey! 8O
If this works I can code this chart into MovieStacker and
make a Prediction Tab to make all the choices and calculations! :wink:

Quote:

Originally Posted by muaddib

Hey! 8O
If this works I can code this chart into MovieStacker and
make a Prediction Tab to make all the choices and calculations! :wink:

Yesssssssssss :mrgreen:

I need people to test this. The chart was based on a 20 second sample. If the prediction is not at least 98% accurate, then I'll do a batch job overnight with a longer sampler, and then feed in the new values to the spreadsheet.

-kwag

Ok, my movie was Ronin.
MovieStacker says average bitrate was 2064 kbps.
Bitrate viewer says 1495
2064/1495 = 1.380602007 x 100 = 138.0602007 ?!?!?!

I used 528x480 with this script:

Code:

Import ("G:\sample avisynth scripts\LoadPlugins.h")



Mpeg2Source("G:\video_ts\Ronin.d2v")



LegalClip()

Telecide()

Decimate()

Gripcrop(528,480)

Gripsize(resizer="lanczosresize")

mergechroma(blur(1.58))

mergeluma(blur(0.2))

SpaceDust()

FluxSmooth(7,7)

NoMoSmooth(40,1,6,1,3,false)

Convolution3d(preset="movieHQ")

DctFilter(1,1,1,1,1,1,.5,0)

GripBorders()

LetterBox(0, 0, 8, 8)

LegalClip()



Sampler(length=24)

What did i do wrong?

Quote:

Originally Posted by Smoochie3

What did i do wrong?

2064 kbps average, 8O, no way :!:
Look again at the LEFT value on MovieStacker.
Right under the "Minutes" field.

-kwag

LOL! :imstupid: i figured it out, thanks!

Ok, so....

Ok, my movie was Ronin.
MovieStacker says average bitrate was 1001 kbps.
Bitrate viewer says 1495
1001/1495 = .6695652174 x 100 = 66.95652174.

So, i would use a CQ of 83?

Quote:

Originally Posted by Smoochie3

That's it :)
Tell me your results :wink:

-kwag

8O Wow. A lot has happened since I was last here......

It might be just me but when I ran a full sampler, got the correct CQ, and then used this CQ and ran another sampler using manual file size prediction, the resulting file was about 2 megs bigger than my target size... :(

Quote:

Originally Posted by PyRoMaNiA

WARNING -SLIPPERY WHEN WET :lol: ALPHA ALPHA ALPHA Method, not even BETA

That's related to ALL my posts in this thread.
Ok, so you mean your "sampler" file size was ~2MB larger than with the standard method of file prediction :?:
What was your resolution :?:

-kwag

hey kwag,

I tried the table and it worked great 1/2 times- first time very close- second time off pretty badly :? If you want you can download my excel worksheet of Heist that I posted-- did you see that?

I'm confused about a couple things though:

1. The curve range is totally different between 352x240 and 528x480 (assuming the resolution is causing this)- so won't the error factor be different?

2. Why not just use filesize and not bitrate? Using the bitrate is a serious pain- first the recommended bitrate to use in MovieStacker is based on all the settings on the page (most relate to SVCD) :roll: It would be so much easier to just enter your error factor into the Predict2 function and it will display the correct cq to use :idea:

thanks,
ren

Quote:

Ok, so you mean your "sampler" file size was ~2MB larger than with the standard method of file prediction :?:

That's right, and my resolution was 352x480.

Quote:

Originally Posted by rendalunit

Hi ren,

Yes, the resolutions throw off the curve completely 8O.
I guess this exercise did turn into a "crap" ( for my posts :cry: ), because the method is very tightly integrated with the MIN, MAX and resolution I used. So there would have to be a chart for every template, and then if anything is changed ( matrix, etc. ) that would mean that a whole new chart would be needed. So, I slipped and fell on this one :twisted:
Thanks for trying it out. Now I'll go dig a hole in the ground and proceed to insert myself in it for a couple of hours, and meditate :roll: :x :x

-kwag