vbv constraint vs undersize avoidance

[Sharktooth]

just a question, why not making pass2 and altpass2 use the same "scale" for the overflow settings?

[plugh]

Hey squid_80, did you get around to enabling the debug stuff for altpass2 in encraw? (ref here (http://forum.doom9.org/showthread.php?p=915565#post915565))

If anyone ever posts a problem, I/they will need to be able to turn that on for diagnosis...

[plugh]

just a question, why not making pass2 and altpass2 use the same "scale" for the overflow settings?

Uh, I think the answer to that is in the posts I ref'd.

1) the original behaviour was not ideal to begin with; over-correction was visible in frame size plots.

2) I changed the underlying quant calculations, which resulted in even less strength being required.

3) Quote: "I have neither the test material, the time, nor the inclination to do the test encodes necessary to re-tune the file scale factor shift points and overflow boost multipliers in that bit of code." - we are talking lots and lots of encodes and statistical analyses of the results (at least that's how I'd tackle it)

And finally 4), at least for the material I've tried, values in the 5/5/5 to 10/10/10 range seem to do just fine without "smart overflow setting" enabled.

If/when I get feedback that demonstrates a need for further strengthening as a function of the file size ratios, I have NO issues with turning that back on with adjusted values - once some 'real world' cases give some idea as to what those values should be. (It would be nice if whoever put that bit of code in there in the first place would step forward and comment on how the original values were derived. I was not able to find any info in the xvid devel mailing list archives).

In the meantime, the current behaviour is actually consistent with both the comments in the original code and the description of the parameters in the 'Unofficial Xvid FAQ' - which it WASN'T before. The only somewhat siginificant departure is the Zero cases - and I personally find great value in being able to turn that stuff off, which was impossible to do before.

Hey, this rework of the pass2 code is experimental. It appears it is getting used and discussed (I did some googles recently and was amazed at what turned up) and so far it seems to be working ok for people, which is encouraging. It sort of motivates me to finish off the other parts of the rework and eventually submit the whole thing (both the pass2 and vfw bits).

Anyway, 'thanks' for your interest and I hope I answered your questions...

[Sharktooth]

yes,thank you.

[plugh]

Thank *you* - with this being supported in MeGui it will get more usage, and either reveal any issues or provide further confidence in the rework.

And to reiterate: if you currently have '0' for any overflow parameter, change it to '10' and you'll get the same result as '0' when they are using pass2, and will avoid the 'turn off' feature if they are using altpass2. And 5/5/5 is an 'ok' set of values to use with both, though it might need to be bumped up a little with some altpass2 encodes.

[Sharktooth]

thanks for the tips. ill get it working for the weekend.

[Sharktooth]

well... in the last post i was a bit optimistic... i couldnt know... but i had serious health problems that prevented me to do what i planned.
however, better late than ever...
im about to commit DivX profiles support for xvid... so... i hope you're still interested in that...

EDIT: the new megui version supports DXN HomeTheatre, DXN HD 720p and DXN HD 1080p profiles. More to come...
Custom profile support is also coming...

EDIT2: ...oops... i made a mistake... i'll correct it tomorrow...

EDIT3: fixed. new version is up.