PDA

View Full Version : Overflow improvement/degradation settings


Prettz
19th May 2003, 23:23
These two settings simply affect how aggressively/quickly the codec can compensate for oversized and undersized sections, correct?
I understand that theoretically there's always the risk that using high values for these can cause to codec to allocate too many bits to a scene and not have enough left over for later ones, or to not allocate enough for a difficult scene that appears suddenly after one with no motion. But would anyone care to give a more concrete discussion on the subject, i.e. at what point with the values would you begin to enter the "danger zone", and how effective is xvid at being able to handle aggressive values and still make the best overall judgement on bitrate allocation?

I had usually left the values for both of these settings between 60%-70%, but I heard someone mention a few months ago that he got good results on anime using the alt-curve mode with the "aggressive" setting. I decided to try encoding an anime (still using the normal curve compression) with values above 75% (and most recently with both at 80%), and the results seemed to be better than with values of 70%. Are very high values generally a good idea with anime (including high quality ones with plenty of CG)? I was particularly thinking this would be good if you have a FILM anime where in some scenes the actual animation isn't even the full 24fps (hence the niche for Donald Graft's Dup filter).

OUTPinged_
20th May 2003, 22:57
Actually those values are only usefull for really small frames - big positive/negative overflow values were hitting them too hard and the quality was unacceptable.

Higher values will hurt small frames.

Lower values will reduce filesize precision.

I believe that you cannot improve quality noticably by tweaking this one. Default works very well (i checked it). Also it is absolutely unnecessary to tweak it if you dont use modulated (hq) quantization and "canonic" 2pass encoding. - overflow will be too small to make any difference.

Better spend your time messing with mrq and distance values.