Using final avg. quantizer like a compressibility test value

[jonny]

Comp. test value is % value calculated with a sample of the movie, and indicates how the movie is compressible.
If comp. test value is high, final avg. quantizer will be low.
If comp. test value is low, final avg. quantizer is high.
So why don't use the final avg. quantizer obtained by a sample of the movie like a comp. test value?
With this method we can test the full range of DivX 5 options... (like keyframes settings)

It's simple to do this:

- In GKnot save the .avs file with the % of the movie you want to test.
- Encode the .avs in 2-pass mode (with DivX 5 settings you want to test and bitrate given by GKnot)
- Open the final analyse.log, read avg. quantizer and you have the compressibility test!

I know this method is slow... but i think is more accurate for DivX 5

Please tell me what you think! :D

-----
jonny

[dragoman]

Hi,

I think this would not be too accurate.

First, you are supposing that there is a linear relationship between average quantizer and filesize, and if you simply look at the avg quantizer you will be able to predict the filesize for the movie.

This is not so: the codec chooses a quanitzer in two-pass mode in order to fit the desired bitrate. Factors that influence this are:

1) Length of the movie (how many frames)
2) Compressiblility of the movie
3) Resolution chosen

Another thing is that you can't really tell how an entire movie will compress by looking at a short scene of it (hold on, I'll explain)

The compressibility test in Gknot works by choosing a certain number of frames from the entire movie (based on your choosing, usually 5%). So roughly 5 out of every 100 frames in the entire movie are used. This provides a good indication of how the entire movie will be compressed.

What you are talking about is taking a section of the movie all in one chunk and applying the avg quanitzer for that to the entire movie. You can't do this, because very few movies will have the same avg quantizer for all scenes. Some are low-motion, some high.

I hope I've explained it enough....let me know if not.

dragoman

[jonny]

I think you are wrong


>First, you are supposing that there is a linear relationship between average quantizer and filesize, and if you simply look at the avg quantizer you will be able to predict the filesize for the movie.


I'm supposing there is a relationship (not linear) between average quantizer and compressibility value.
(What you say is not true too, there is a relationship between average quantizer and filesize (not linear), if we test the same movie, increasing the filesize (example: increasing the bitrate) actually decrease the final average quantizer.)

I've made a lot of testing with full movies, calculating the compressibility test and storing the final average quant.
I hope to clear this by some example:

Movie 1:
Compressibility test: 53.3%
Final Avg quantizer: 4.286401

Movie 2;
Compressibility test: 61.6%
Final Avg quantizer: 3.934464

Movie 3:
Compressibility test: 69.3%
Final Avg quantizer: 3.597155

Movie 4:
Compressibility test: 81.5%
Final Avg quantizer: 2.918067

What i'm saying is that:
"Final Avg quantizer = 2.918067" is the same to say "compressibility test value = 81.5%" and so on...


>Another thing is that you can't really tell how an entire movie will compress by looking at a short scene of it


- In GKnot save the .avs file with the % of the movie you want to test.
This is equal to say for me: enable comp. check, select the % you want to test and save the .avs ( sorry for not making this clear :) )


I hope to make my concept more clear now... (sorry but my english is poor)

[dragoman]

Hi,

My mistake.....you are correct, upon re-reading your original post I realize I was wrong.

dragoman

[Alex_e_Basta]

Hi. I followed a similar way in order to determine the DRF distribution. I made a small tool that examines analyze.log and produces a diagram with the New Quant distribution.
The average bitrate that I find in the first record of analyze.log is always greater than the sum of Newquant / total frames. I think that avr bitrate is calculated at the end of the first step , so it is not exactly the final avr bitrate .
What do U think about this ? Is New Quant a reliable information ?

TIA Alex.

[jonny]

Ciao!

The test i've made show that there is a relation between average quantizer (in the first line of analyse.log) and compressibility test value.
I'm not sure that the "new quant" values are reported correctly, but i don't know how to test this.
(At this point i'm not sure that the avg. quant in the first line is ok too)

Is there a large difference between average quant reported VS average quant calculated by you?

-----
jonny

[Alex_e_Basta]

Ciao jonny !
Here the calculated difference for the same movie, at the same resolution and with the same one bitrate.
Avr Bitrate from analyze.log; from my tool
Divx4 3.9977; 3.6048
Divx5 3.4611; 2.9173

Same movie, BUT killed after encoding 15700 frames
Divx4 3.9977; 3.5186

As you can see, analyze.log always shows the same avr bitrate
for an interrupted film too,therefore I think it calculates avr Quant before the beginning of second step. In coding phase the codec determines the New Quant in a different way.
The question is : which value is right?

Bye.
Alex

[jonny]

@Alex_e_Basta:

Check your PM in user cp!!!
(sorry for this not useful post, but no other way to contact this guy :D )

[jonny]

@Alex_e_Basta:

Check your PM in user cp!!!
(sorry for this not useful post, but no other way to contact this guy, i think this time i'll ask him the e-mail :D)