View Single Post
Old 28th June 2005, 08:32   #16  |  Link
MOmonster
Registered User
 
Join Date: May 2005
Location: Germany
Posts: 495
@all
Once more, I updated the functions. The bmode 2 no longer needs the unfilter and the functions have a better handling. The modded version no longer has the nlv value, but a very important motion level value (please read the updated text at the beginning).
I added a new configurable parameter to tweak the blend detection, but in general I wouldnīt touch this parameter, called "hv". This parameter is used for the internal calculation ot the blendtest. Useful values are between 0.6 and 1,4. Normally, higher values help to find the blendings better, so for the silent Cdeint you can use for example 1.0 (default is 0.9). For A clean, sharp (not oversharpened) source set smaller parameter. For worse sources use higher values. If you see that there are blends with factore under 20% are left the default value is maybe to less. For the modded version this value should be set more accurate (default is 0.8), because of the updated internal condition function, it have to recognize really clear.

@yaz
Like you can see it, because of the posts in your thread, I also see no way to restore this nicely. Now I know why my function works better for your source (because of the many blends that stays together). But wonīt help you much. If we just take all the clear fields the source look not so bad, a little bit soft, but no strange artefacts because of the compression. It looks like a bad conversation from Pal to Ntsc and another bad conversation back to Pal or somthing like this.
The big probleme, that some clear fields between two blends are totally missed. So to get not a to strange motion, we have to took one of the blends. Ok some blends left wonīt be the probleme I think, but the strange dark stripes, we can see in every blend. This looks to bad. I donīt think that you get better results than using Cdeint with tfm (or to save some more clear fields the new modded version, but this wonīt make the difference) on it.
Maybe sharfis_brain or Didee know what to do against this strange artefacts, but killing more blends wonīt be good for the motion. Only theoretical it is possible to restore this missing clear fields by adding the two blends together and than substract the blend of the two clear fields around. If we exactly know the blendlevel, it is possible to restore also this cases. But sorry this is just to rare and the function really to complex and would only work with really good compressed material, so I think I wonīt try something like this in near future. Maybe it is the best to buy the ntsc version. I donīt believe that it will also be that bad.

Last edited by MOmonster; 25th July 2005 at 23:45.
MOmonster is offline   Reply With Quote