[XviD-devel] Forced iframes lost?

[Dirk Knop]

Hi,

over at doom9 we fell over a small glitch which results in unneccessary 
"bad" encodes:

We take the iframes from the first pass (-> stats file) and force them 
in the second pass.
But reality looks different:

*1st-pass*
Number Of Intra-Frames (Key-Frames) : 180
Number Of Inter-Frames (P-Frames)   : 6476
Total Number Of Frames              : 6656

*2nd-pass*
Number Of Intra-Frames (Key-Frames) : 173
Number Of Inter-Frames (P-Frames)   : 6483
Total Number Of Frames              : 6656

I looked through VFW and couldn't find any false "assigns" there, so I 
think this is an issue from the core.

This results in some weird things with overflow: at sch a false 
non-keyframe the overflow jumps to a very high value (well, actually a 
very low value like -20000 bytes), which then gets redistributed as 
gracefully as possible, but still resulting in quants +1, +2, +3... 
where without this "feature" we would continue with the average 
quantizer we had before.

Ah, this is concerning the "stable" tree btw., with dev-api-3 2pass is 
totally messed up due to the dynamic I/P/B decision mechanism ;)

Anyone with more insights into this matter?

Best regards,
Koepi