Undersized

[BiaTch 5.0]

I don't know if it's a GKNOT bug or a problem with divx 5.02 but when encoding Jeepers Creepers I got 44mb (656mb)undersized.

704x384 (full)

83% comp test

[OvERaCiD23]

When a comp-test returns a value above 80%, an under-sized file is likely. The movie is getting saturated, meaning it's at full quality. By searching, you would find that there are remedies for this. Use a sharper resize, increase audio bitrate (or retain the original AC3). Dark movies with little motion are very compressible, and 'Jeepers Creepers' is just that.

[jonny]

@OvERaCiD23:
Only one correction: the movie don't get saturated at 83%, it's a codec bug if the result is undersized.
A good working codec should saturate at comp.test above 100%

[jggimi]

A good source of info on undersized DivX can be found here (http://forum.doom9.org/showthread.php?s=&threadid=24584).

[Mac Sidewinder]

A good working codec should saturate at comp.test above 100%

It should but I have gotten undersized files when the comp check was between 80 and 90%. Either you live with it or you simply try some of the ideas that jggimi pointed out.

Mac

[jonny]

My point is that at 83%, the resulting undersized file is not at maximum quality = not all the quants are equal to 2.
The only method to force the codec to break this limit is to switch to quality based mode.
Is this info so bad?

[BiaTch 5.0]

Thanks for the help guys

[N_F]

Originally posted by jonny
The only method to force the codec to break this limit is to switch to quality based mode.


Not quite. Let's say you get a 10% undersized file with 1000 kbps. If you increase the bitrate to 1100 you'll probably end up with a 2-5% undersized file (depending on original compressability). If you then increase it to say 1180 you may well reach your desired filesize (assuming it doesn't have a comp. value of >100% of course)

This has at least been my experience with DivX 5.02. (Perhaps you already knew this but just didn't want to go into detail, don't mean to nitpick)

[jonny]

I was missing this, thanks N_F, you are right.
On my tests with 5.0.2 another thing that help is to lower the maximum quant to 4 or 3, depending on comp.test value ( 3 can be a little dangerous, sometime i've obtained oversized results :rolleyes: )