Bitrates: Testing CQmatic Versions - Page 4

Krassi · #61 07-31-2003, 02:17 AM

Here are some movie details (can't figure out if its interlaced because i don't have access to the source right now) but here are some movie details:

Code:

Stream type: MPEG-2  MP@ML  VBR
Resolution: 704*576
Aspect ratio: 16:9 Generic  
Framerate: 25.00
Nom. bitrate: 3137200 Bit/Sec
VBV buffer size: 112 
Constrained param. flag: No 
Chroma format: 4:2:0
DCT precision: 10
Pic. structure: Frame
Field topfirst: Yes
DCT type: Frame
Quantscale: Nonlinear
Scan type: Alternate
Frame type: Progressive

AFAIK its not interlaced, only some tv-stations are having interlaced material in germany

kwag · #62 07-31-2003, 02:30 AM

Thanks Krassi,

I asked, because I'm having problems prediction interlaced material. That's what I'll be tackling tomorrow (actually today

)

-kwag

jorel · #63 07-31-2003, 02:56 PM

only removing doubts:
my source is 29,97!
the first fast search in CQMatic 1.03 using CQ50 encode as 0,0023976!
is ok?

kwag · #64 07-31-2003, 03:05 PM

Quote:

Originally Posted by jorel

only removing doubts:
my source is 29,97!
the first fast search in CQMatic 1.03 using CQ50 encode as 0,0023976!
is ok?

Yes

-kwag

jorel · #65 07-31-2003, 03:16 PM

Quote:

Originally Posted by kwag

Quote:

Originally Posted by jorel

only removing doubts:
my source is 29,97!
the first fast search in CQMatic 1.03 using CQ50 encode as 0,0023976!
is ok?

Yes

-kwag

thanks!

kwag · #66 08-06-2003, 09:54 AM

Here's the result of my first full encode with CQMatic version 1.1.02

Movie: K19 (138 minutes)
CQ found: 56.64
Wanted file size (Moviestacker) : 699,875.78KB
Final encoded file size: 697,860KB

Using this raw script:

Code:

## DLL Section ## 
# 
#LoadPlugin("C:\Filters25\MPEG2Dec3.dll") 


Mpeg2Source("F:\K19\VIDEO_TS\k19.d2v" )
# 
BicubicResize(528, 272, 0, 0.6, 5, 57, 710, 366)
AddBorders(0, 104, 0, 104)
LetterBox(16, 16, 16, 16)

-kwag

Bchteam · #67 08-06-2003, 01:31 PM

The 1.1.0.2 Version is giving me serious problems. On my the first prediction for the 90 Minute Movie "May" the CQ went up to 90. But I encoded the Movie one day before with a CQ of 63,75, which gave a final size, that was 16 MB off target, so 90 just cannot be serious. And another Movie called "2009 - Lost Memories" was a total flop. On that Movie, CQmatic 1.1.0.1 gave me CQ of 44,94, but CQM 1.1.0.2 was going below 30, when I cancelled the whole process. The strange thing is, that the Low Fence value always remains 2.000000, even after several prediction cycles the low fence remains at 2. On both Predictions I did not override the CQ.

Here's my Log for that Movie. In the End I cancelled it, because I saw that it wasn't going to be successful.

Code:

http://www.kvcd.net
CQMatic Version 1.1.02
Copyright Softronex Corporation, 2003.
All rights reserved.
Time: 19:35:48 Date: 08/06/2003
Ready!
Project: G:\Programme\CQ - Matic\Projects\2009Film.tpr

Creating: CQMatic.tpr

G:\2009Film.m1v
Execute.
Movie Time: 136
Average Bitrate: 691
Prediction Only mode
Executing Prediction Phase...
Process started at 19:36:19
On 08/06/2003
CQ set for prediction
Interlaced Source Correction
Setting up initial sampling.
Using CQ of 60.00
Prediction cycle #1
Encoder started...
Process time: 16.43 minutes.
Encoder end.
File size difference  = 0.633915
Low fence: 2.000000
High fence: 60.000000
Last CQ = 60.00
Current CQ = 38.03
CQ difference = 21.965122
Using CQ of 38.03
Prediction cycle #2
Encoder started...
Process time: 16.55 minutes.
Encoder end.
File size difference  = 0.776140
Low fence: 2.000000
High fence: 38.034878
Last CQ = 38.03
Current CQ = 29.52
CQ difference = 8.514488
Using CQ of 29.52
Prediction cycle #3
Encoder started...
Process time: 16.75 minutes.
Encoder end.
File size difference  = 0.834189
Low fence: 2.000000
High fence: 29.520390
Last CQ = 29.52
Current CQ = 24.63
CQ difference = 4.894800
Using CQ of 24.63
Prediction cycle #4
Encoder started...
Process canceled

kwag · #68 08-06-2003, 01:45 PM

Hi Bchteam,

What resolution are you encoding

I ask, because ALL the tests I've done, have been KVCDx3 (528x480).
Now that you mention that, I haven't tested any other resolution

Maybe this is a cause of prediction problems

I'm going to try a 352x480 now, and see what results I get.
As far as 528x480 (NTSC), I'm right on target

-kwag

totonho03 · #69 08-06-2003, 01:49 PM

Hello Kwag:

I am presently testing the new version of CQmatic. These were my steps:
1.- Loaded the new 1.1.02 version
2.- Used 120 minutes with the new average bitrate (779 from Moviestacker)
3.- Using yesterday's .tpr file (From Tmpgenc) Did not change the max or min here. (max 2,000, min 537)
4.- I did not reset TMPGenc at all, therefore it used yesterday's settings.

So far it is in its 5th prediction cycle (Each cycle is around 27 minutes). The CQs that have/are being run are:
60 > 69.73 > 75.83 > 78.07 > 80.26...and counting.
Max bit rate= 2000; min bit rate= 537
non-interlace
Motion estimate search
Average bit rate at CQmatic=779, which corresponds to the new 120 minutes setting.

I was expecting to see a CQ lower than the one obtained yesterday, which was +/- 63, but as you can see, it presently is in its CQ 80.26 run, which means that the file size is going to be way over the 700 Meg target.

Thanks
Totonho03

Bchteam · #70 08-06-2003, 01:50 PM

"May" is 704x480@25fps and "2009 - Lost Memories" is 528x480@25fps. It doesn't seem, that it's resolution related problem.

totonho03 · #71 08-06-2003, 01:52 PM

P.S.- Maybe I should try, for testing purposes, your raw script.

Totonho03

kwag · #72 08-06-2003, 01:56 PM

@totonho03,

Are you encoding 528x480

or some other resolution

This doesn't make sense AT ALL

How can I get a movie to within ~2MB difference (K19), and everyone else CQs are so way off

Something in the encoding chain is WAY different from what I have to what everyone else having problems has. I can run 10 times CQMatic against that same movie, and get the same CQ every time. So I need more information, as to what resolutions are being used, if different that 528x480. I just started a prediction test on a 352x480, to see if there's a difference. I just can't think of anything else right now that can possibly create such drastic differences in CQ

:banghead: :banghead: :banghead: :banghead: :banghead: :banghead:

Thanks,
-kwag

nicksteel · #73 08-06-2003, 02:24 PM

I think that to test CQMatic, everyone is going to have to insure that the other involved programs and factors are consistant across users. For example, everyone should use the same bitrate calculator, or better still, the manual average bitrate formula. Everyone should configure TMPGEnc the same way, i.e., minimum bitrate padding, etc. This way, we can make many tests against many films with CQMatic the only real variable. It would probably be best if we all used the same resolution, fps, etc, for each set of tests.

We could all test 528x480 at 24fps and list results. Then move on to another resolution, fps, etc. It really doesn't matter if the output is usable at this stage, just that the CQMatic generated file is near the desired file size.

If any problems occur, we could isolate them to a particular film or conditions. We need to be giving Kwag consistant test results based upon known factors.

audi2honda · #74 08-06-2003, 02:25 PM

Hi kwag with the latest version my reference DVD jumps straight to a CQ of 90. Using resolution of 704x480, 1686 avg bit rate, movie length 120 minutes.

totonho03 · #75 08-06-2003, 02:26 PM

Hello Kwag:
These are my TMPGenc settings:

Quote:

Video tab

Quote:

Stream type: MPEG-1 Video
Size: 352 X 480
Aspect ratio: 16.9 525 (NTSC)
Frame rate : 29.97
Rate control mode: Constant quality (CQ); Max bit rate 2000, Min 537
Main Profile & Main level (MP@ML)
Encode mode: Non interlaced
YUV Format: 4:2:0
DC component precision: 8 bits

Quote:

Advanced tab

Quote:

Video source tab: Non interlace (progressive)
Fiel order Botton field first (Field B)
Source aspect ratio: 16:9 525 line (NTSC)
Video arrange method Full scren keep aspect ratio
320 x 2440 pixels
The Source range box is presently checked

Quote:

GOP Structure

Quote:

PBBPBBPBBPBBPBBPBBPBBPBBPBBPBBPBB........
1
5823
2
1
24

Quote:

Quantize matrix

Quote:

For intrablock 8 is highlited
For non intrablock 16 is highilited

CQmatic is still running, and presently is doing CQ 80.87
I feel like

for my lack of experience here, but please do not

, but if my failures keep on happening, then they are going to drive me to

cachazas (capirinhas)

. Perhaps I should already hit for the refrigerator and star fermented cebada juice.

Totonho03

kwag · #76 08-06-2003, 02:34 PM

Hi nicksteel,

Yes, you're right

But now I have a new piece of information for everyone, which is really weird

I just did a 352x480 prediction on the movie "Red Planet".
I used a MAX of 2,000 and MIN of (.57 * 919(avg.)) = 523.83Kbps
I got a CQ of 80.25
Now I changed the MIN to 300 and the MAX to 2,500 and ran prediction again.
Surprise

found CQ of 74

Now, which is the correct CQ

Is it the resolution

Is it the MIN

Is it the MAX

Too many factors, and this is getting wild

The only thing I can think of, is that at higher resolutions, the MIN and MAX are being used to the full extent by the encoder, but on lower resolutions, the bitrate MIN/MAX ratio is not being reached during prediction. And that's why there's such a difference from 528x480 to 352x480

This brings me to another question: Which is the correct MIN/MAX to use, if there's such a large difference of CQ with a change of resolution

Too many factors, too many variables, and this is the root of the problem

-kwag

kwag · #77 08-06-2003, 02:40 PM

Quote:

Originally Posted by totonho03

Perhaps I should already hit for the refrigerator and star fermented cebada juice.

Perhaps I should go out running as fast as I can into a bar, and get really drunk

Maybe after drinking a six pack of beers I can think of something, other than hanging the creator of TMPEG for not making the CQ curve linear

-kwag

totonho03 · #78 08-06-2003, 02:51 PM

Sr Kwag:

If you do not mind my 2c., I also believe that nicksteel is correct in his assertion and recommendation. Perhaps what we all need to do is to use the same parameters, then compare notes and see if the different OS and machines, are given us the same results. Once this is established then groups of 2 or 3 can test different parameters. Needless to say that you will provide us with the first parameters/settings and guidance, as well as with the following testing settings.

Regards

Totonho03

PS.- Not that it makes any difference now, but CQmatic just finished its prediction and is given me a CQ of 80.87. It took about 180 minutes to do so.......

kwag · #79 08-06-2003, 02:56 PM

Ok, let me finish some tests at different resolutions to see the results. I want to test the relationship between bitrate/resolutions before moving on.

-kwag

jorel · #80 08-06-2003, 04:06 PM

Quote:

Originally Posted by nicksteel

I think that to test CQMatic, everyone is going to have to insure that the other involved programs and factors are consistant across users. For example, everyone should use the same bitrate calculator, or better still, the manual average bitrate formula. Everyone should configure TMPGEnc the same way, i.e., minimum bitrate padding, etc. This way, we can make many tests against many films with CQMatic the only real varable. It would probably be best if we all used the same resolution, fps, etc, for each set of tests.

We could all test 528x480 at 24fps and list results. Then move on to another resolution, fps, etc. It really doesn't matter if the output is usable at this stage, just that the CQMatic generated file is near the desired file size.

If any problems occur, we could isolate them to a particular film or conditions. We need to be giving Kwag consistant test results based upon known factors.

great observation my friend nicksteel,very logical.
i was thinking the same but i can't be so clever.

Kwag wrote:
"Too many factors, too many variables, and this is the root of the problem

"
"

"Too many factors, and this is getting wild

"
i post it somewhere here: if i encrease the "MIN",and decrease the "MAX",
the CQ encrease and the final size encrease too

Otto
give me 2 "caipirinhas",i need it too!