CCE OPV prediction failed due to too high max_bitrate

[pilar]

Hi,

Just tried to test the OPV in CCE trial 2.70.2.4 on Spirited Away R2, and RB v1.06.1 failed at the prediction stage:
-- Sampling 3588 of 179400 frames.
-- Predicted size (sectors) at Q=24: 2 118 626
-- Predicted size (sectors) at Q=27: 1 988 132
-- Predicted size (sectors) at Q=28: 1 222 465
-- Predicted size (sectors) at Q=13: 2 735 033

Evidently, something went wrong at Q=28. I've copied analysis.ecl and avs files to different directory, aborted the prepare phase and tried to ran CCE against those files again. CCE failed with VBV error leaving partial m2v, and that probably fooled RB prediction analysis.
I tried again with lowered max bitrate from 9000 to 8800 and this time CCE finished succesfully. Then I noticed that 9000 max bitrate was set in analysis.ecl, but in rebuilder.ecl max bitrate is different for each segment, with highest value about 8400.
It seems to me, that the max value from rebuilder.ecl should be set in analysis.ecl. In my case this would prevent the error and allow to predict Q value correctly.

Edit: Forgot to include the important line:

-- TargetSize (sectors):1 927 641

[jdobbs]

That's really strange... I'll look at it.

[jdobbs]

It seems to me, that the max value from rebuilder.ecl should be set in analysis.ecl.That really doesn't matter for prediction -- the size of the output is determined by the average bitrate setting, not the maximum.

[pilar]

That really doesn't matter for prediction -- the size of the output is determind by the average bitrate setting, not the maximum.

Don't forget, that we are talking about the OPV - average bitrate doesn't apply here, but max bitrate do. Size here is determined mainly by Q value. Just out of curiosity, I tried to run the CCE against the analysis.ecl in above case at max bitrate 8800 and 8000, and the difference in m2v size was about 1MB, that means 50MB for whole dvd.
I agree, that's small enough to don't care about, and that wasn't the point of the phrase you quoted.
What I meant was that maybe there's some kind of threshold (or maybe bug) in CCE that cause its crash with VBV error when max bitrate is too high.
I didn't see something like that reported, so it's probably very rare case. But, as I don't see the reason why the max bitrate in analysis.ecl should be higher then in rebuilder.ecl, I advise to make it the same as max there. I think, it shouldn't hurt anything.

[jdobbs]

The target size for OPV is determined by the average bitrate. So the average bitrate will determine the Q value selected.

As I said, and as is still true, the value for maximum bitrate has no bearing... ;)

[pilar]

The target size for OPV is determined by the average bitrate. So the average bitrate will determine the Q value selected.

Oh yes, I see. I thought you are talking about the average bitrate setting in CCE.

Just for completness, here is the exact CCE error that occur at max_bitrate=9000:
cce encoding failed: OPV VBV ovf frame# 2316 (00:01:32:18) I 100628 max 100161,44 rel 466,56 1,00 qsv 7,68->7,77

[jptheripper]

now you got me curious, and i am totally ignorant to cce errors.. however

you said 8400 worked but 9000 failed, if it is easy could you run 8500 and post the error? it would be interesting for me to see the difference in the error messages.

[jdobbs]

Also -- look in the analysis.ecl and see what Q value is being used. I know it says 13 in the log -- but I want to be sure.

[pilar]

@jptheripper
It worked even for 8900. Here is the error message for 8950:
cce encoding failed: OPV VBV ovf frame# 2316 (00:01:32:18) I 99816 max 99536,38 rel 279,62 1,00 qsv 7,79->7,88

@jdobbs
The prediction in RB failed at Q=28, and that's the value I used for every further test (it also happened to be the right value for backup and after I set that Q manually in rebuilder.ecl, RB finished encoding and rebuilding successfully).

[jdobbs]

Hmmm.... I may have to eat my words about the maximum bitrate. While I still contend it would have no impact on the prediction... it apparently does have an impact on the crashing. :angry:

I tell you -- I've sure had to write a lot of "special" code for CCE v2.70... it seems to be very fickle.

[pilar]

While I still contend it would have no impact on the prediction... it apparently does have an impact on the crashing.
That's what I was trying to say.

I tell you -- I've sure had to write a lot of "special" code for CCE v2.70... it seems to be very fickle.
You are right. Just out of curiosity I've checked the same test with CCE 2.67 trial and no problem here.

[jptheripper]

sorry jdobbs, that stinks.

pilar - very nice beta testing!