Different compressibility checks for the same movie ?

[iparout]

Hi.

I was encoding a movie the other day (can't remember which one) and then a friend of mine borrowed my DVD in order to encode it himsalf as well. When we both did compressibility checks for the movie, it turned out that I got around 58% and my friend got somehing like 40% !!! The weirdest of all is that the we bith used the *EXACT* same procedure when ripping the movie and furthermore the GKnot settings for the compressibility check were exactly the same as well...

The only difference is that he has a Pentium III 800 MHz whereas I got a Pentium 4 2 GHz.

We also used other movies to see if the difference in compress. checks could occur again and it did ! For every movie, although the DVD + ripping procedure are exactly the same, he gets around 15% lower compressibility check than mine.

Can anyone figure out why this happens ?

Thanks in advance.

[jonny]

Are you sure you are using the same version for avisynth+plugins, virtualdub versions ecc...?

[iparout]

Originally posted by jonny
Are you sure you are using the same version for avisynth+plugins, virtualdub versions ecc...?

Now that I am thinking of it, I am using the latest version of Avisynth, whereas he is using the one which came with GKnot 0.26. We are both still using the 0.26 GKnot. Does the avisynth version make such a difference ? I mean, can it improve compressibility by that much ?

[jonny]

There are to many factors to be considered...
Please check you are using exactly the same softwares... parameters, filters ecc...

[manono]

Hi-

The resizer alone could account for the difference (most of it anyway), or the audio bit rate, or the file size you set. But I'd compare the 2 .avs's first.

I don't think the AviSynth version would make much difference, but the Plugins used certainly could.

[iparout]

I am 99% sure that *ALL* features of the cempressibility check were the same. I'll try to post the .avs of two compressibility checks when I get the time. It may take long cause I don't have time right now, but I'll et back to you.

Thanks for the help so far.

[OvERaCiD23]

The % of the check could amount to that difference. If you did a 5% check and he did a 1%, there could be lots of variance. Using the exact same settings, you should have results within ~1% of one another (CompChecks vary a little in my experience, sometimes even if I run the same one back to back).

[jonny]

(CompChecks vary a little in my experience, sometimes even if I run the same one back to back).


Are you sure about this?... sounds strange for me (never happen with divx5enc, i obtain always identical results)

[bkam]

Yeah, like Jonny said, the difference I've had between compressibility tests of 1%, 5%, and 10% is never more than ~2% or so as I recall. I haven't done extensive testing but I used DivX5Enc to play around with it a few times and at least for the movies I tried they all gave almost exact same values. I guess in theory a low % could "miss" a lot of hard to compress parts, but it seems unlikely to me, I mean, even a 1% test checks 14 frames every 1400 frames, I think, which is every 58 seconds. It seems like harder-to-compress parts, unless they were spaced at a ridiculously even interval that somehow the comp test missed, would show up enough to give a pretty good idea how compressible the movie is. I mean, a movie isn't gonna be 15% less compressible because of a few 30 second parts that are hard to compress--consider that it's testing more frequently than every minute even at 1% (Unless I am totally wrong about how comp checks work). At 5% it checks some frames about every 12 seconds of film I think, of course this normally be slightly better, but how often in movies are the scenes changing so fast that you don't even spend a minute in one of them? At least that is my take on it, I'm not sure if I'm totally wrong.

[iparout]

Originally posted by OvERaCiD23
The % of the check could amount to that difference. If you did a 5% check and he did a 1%, there could be lots of variance. Using the exact same settings, you should have results within ~1% of one another (CompChecks vary a little in my experience, sometimes even if I run the same one back to back).

I know that, that's why we both did a 5% compressibility check. Eventhough, according to my experience, the difference between 1% and 5% commpressibility checks is never more than 5-6%, let alone 18-20% as the case is in my situation.

[OvERaCiD23]

Are you sure about this?... sounds strange for me (never happen with divx5enc, i obtain always identical results)

Positive. I can run the 1st pass in VDub, then after it's done run it again (not changing a thing), and can vary a little bit. It's never been more than a 1% difference, but still. I'm not sure about it, I just write it off.

No 20% difference is quite large, I didn't mean it could vary that much. ~5% is more common between the different length CompTests.

[jonny]

Positive. I can run the 1st pass in VDub, then after it's done run it again (not changing a thing), and can vary a little bit.


Sorry but i'm in disappoint with this.
I'd run a series of tests last week using divx5enc, and the predicted size at 1% was 1,223,018,966 bytes.
This week i have redone 1% to verify the parameters: predicted size was 1,223,018,966 bytes again.

Simply put 2 identical jobs (at 1% for example) in divx5enc, and you'll obtain always the same result... please, make this test :)

cheers
jonny

[jonny]

@bkam:

>I guess in theory a low % could "miss" a lot of hard to compress parts, but it seems unlikely to me.

In some cases this happens, but usually you obtain bad results only at 1%

>I mean, even a 1% test checks 14 frames every 1400 frames, I think, which is every 58 seconds. It seems like harder-to-compress parts, unless they were spaced at a ridiculously even interval that somehow the comp test missed, would show up enough to give a pretty good idea how compressible the movie is. I mean, a movie isn't gonna be 15% less compressible because of a few 30 second parts that are hard to compress--consider that it's testing more frequently than every minute even at 1% (Unless I am totally wrong about how comp checks work).

You are right.
I can tell you more, the current divx5enc version (on my HD) have parametric snipsize, i'm testing snipsize=4 (so 4 frames every 400 for 1%) because i think that reducing the interval give a lower error.

>At 5% it checks some frames about every 12 seconds of film I think, of course this normally be slightly better, but how often in movies are the scenes changing so fast that you don't even spend a minute in one of them? At least that is my take on it, I'm not sure if I'm totally wrong.

Yes, with 5% you usually obtain very good results

Cheers
jonny

[N_F]

@iparout

Those percentage values are those one could expect if one of you had B-frames enabled and the other didn't. It can easily be overlooked since it's in the options tab where you don't look very often.

Any chance this could be the case?

[iparout]

Originally posted by N_F
@iparout

Those percentage values are those one could expect if one of you had B-frames enabled and the other didn't. It can easily be overlooked since it's in the options tab where you don't look very often.

Any chance this could be the case?

B-Frames cannot be enabled or disabled for the compressibility check, because this option appears when you hit the "Save & Encode button", in the DivX 5 parameters. The Compressibility Check button is before "Save and Encode".

[N_F]

Yes they can. Go to the options tab and you'll see down to your right some checkboxes. IIRC the box "Use DivX 5 pro features" is unchecked as default, so it won't use B-frames (or any other pro feature) when you do a comp. check unless you've changed this.

[iparout]

Originally posted by N_F
Yes they can. Go to the options tab and you'll see down to your right some checkboxes. IIRC the box "Use DivX 5 pro features" is unchecked as default, so it won't use B-frames (or any other pro feature) when you do a comp. check unless you've changed this.

You beat me to it !!! I just though of it myself and I told him to enable the Pro Features under the Options tab of GKnot. He wasn't using them for the compressibility check. I'll have the result by the end of the day and I'll keep you posted... However I think that this is the solution.

Thanks a lot.

[N_F]

Glad to be of help.