1-pass VBR in HC

[mikenadia]

@ hank315

Got more info on the 10 031 frames clip undersizing issue
http://forum.doom9.org/showthread.php?p=1375087#post1375087

Undersized with AUTOGOP 15, GOP 15 2, GOP 15 1
Right on target with GOP 15 0,  GOP 12 0
In-between for AUTOGOP 12, GOP 12 2 , GOP 12 1
All the other settings in INI do not seems to be involved ( Yes or No ) do not change the issue.

At 50 % of the encode , the average bitrate is much closer to the targeted bitrate ( and to the 2-pass bitrate at 50 %)  with GOP 15 0 and GOP 12 0.

Thanks in advance.

[hank315]

Hard to tell what's going wrong, seems the estimated Qvalue is way too high.
What is the basic Qvalue ?
Is it possible to upload the source ?

[mikenadia]

The base Q is 7.27. Will try to upload the source. It is a Trim of a commercial DVD. So I have to cut it and be sure I can replicate on the new VOB. Thanks

Edit: on that file (62 MB) , I use Trim(10,2031) . It seems to be frame sensitive.
http://www.sendspace.com/file/0ti4mw

Same file ; demuxed (46 MB)
http://www.sendspace.com/file/9boqv3

Base Q:4.61
targeted bitrate: 4000 kbps matrix: MPEG
GOP 15 0:       3916 kbps (on target)
AUTOGOP 15 :  3701 kbps


The 10031 frames clip could be uploaded if needed .

Base Q:7.27
targeted bitrate: 4000 kbps matrix: MPEG
GOP 15 0:       4004 kbps (on target)
AUTOGOP 15 :  3616 kbps

[hank315]

This clip will do, makes a nice test clip.
The 2 pass encode is exactly on par but the 1 pass VBR allocates not enough bitrate to the first part.
And it is, as you said, very sensitive for the frame range.
This requires some serious thinking...

It's always the same with those french movies, too difficult 

[mikenadia]

I agree for the french movies. Tough to encode and tough to understand. I won't buy them anymore.
GOP 2 1 , on the above-mentioned clip, is almost on target ( and gets worse as GOP length is increased) either because of the number of B-frames or because it affects base Q.

CQ 5 with AUTOGOP 15 and GOP 15 0 leads to different bitrate but 1-pass VBR leads to identical Base Q.

Theoretical question: If I encode at CQ=10 (no interpolation)  and encode at that bitrate at 1-pass VBR, should the base Q be 10 or not .
Did a 1-pass VBR with DVD-RB . Very impressive.

Thanks

Edit: Because you liked the movie so much, here is another clip.
Issue previously reported here (but did not know about it)
http://forum.doom9.org/showthread.php?p=1147781#post1147781

Using the GUI;(max bitrate:8000 kbps) : average bitrate at CQ=2 is lower with matrix  Fox3 ( 6979 kbps) than matrix MPEG (7445 kbps). Explanation below . CQ=3 looks OK ( Fox 3 in the 7300 kbps and MPEG in the 6800 kbps ).
http://www.sendspace.com/file/jiuc4c
http://www.sendspace.com/file/wd6fnc  (demuxed)

It looks it is due to the high average bitrate of the sequence and the quantizer of the I-frame cannot stay constant (too close to max bitrate of 8000 kbps); for CQ=2 the average quantizer is much higher with Fox3 than MPEG (that explains the relationship between bitrates at CQ=2).
For Fox3, the average quantizer is 4.72 at CQ=2 and 4.52 (smaller) at CQ=3 ( with iP_factor=1 and GOP 15 0 but similar relationship with AUTOGOP).For the interpolation of Base Q, I have no clue if one should take Q=2 or Q=4.72 ( or average of I-frame quantizer).
Or go with CQ=4,7,10 (too close to CQ=2 seems to be too much affected by max bitrate).

Edit2:Really cosmetics
a) may be start GOP length at 2 (crashes at 1)
b) CQ and VBV/birate unchecked gives in the window CQ=0 (computation is made at the appropriate CQ)
c) CQ and 1-pass allowed simultaneously in GUI
d) it seems that the info message (" Very high quantizer " for example) does not appear on the log. Would be nice when batch encoding.May be base Q in the log?

Edit3: Base Q does not take into account max bitrate.
Same Base Q (5.78) for a small clip with Fox1 targeted bitrate=6 000 kbps with max bitrate 8 000 kbps or 28 000 kbps ( I know not DVD-compliant: just to show my point).
If I use GOP 15 0 , I have " right on target " with max bitrate= 28 000 kbps but 4 980 kbps with max bitrate=8 000 kbps. Same behavior ( with much less differences) at targeted bitrate of 4 000 kbps.
The file used is the one in this post.

It looks like the relationship between size and CQ (  for CQ <4  or close to the max bitrate) depends upon the compressibility of the source.Some information about it can be gathered by doing CQ=5 and 10 before. A better approximation of the behavior close to the max bitrate can be guessed for CQ=2.