View Single Post
Old 25th December 2002, 19:07   #53  |  Link
OUTPinged_
MooPolice 1st division
 
OUTPinged_'s Avatar
 
Join Date: Dec 2001
Location: VIlnius,LT
Posts: 448
[edit] the test was ready yesterday and i dont know if i would do it after several last posts... oh well, hope it will be interesting for some people to check [/edit]


I added one more contestant there, force film with interpolated FD over it. That way you we would see what we get in terms of speed.

4 test samples were used. First was the film content. Second was half-film with "evil" scenechange and half-interlaced. Third was the "non-ivtc-able" one, the stuff you see on NGE, Ranma and other old tape-source dvds. Fourth was the "mouth" clip.

As ivtc22 and FF were leaving interlaced frames slipping on 2 and 3rd clip, i had to use deinterlacer for them. I checked with both blendfields and FD in interpolate mode.

More: I did some tests and it seemed the only difference between ivtc22's 5 settings suggested (for given 4 clips) was the mouth clip working. I have only one of them and I cant tell if those settings are any better or not. All settings worked exactly same for other clips, so 60,11,110 was used. No deinterlacing was required for "film" clip, so it wasnt there. Thus the speeds.

As I value my time YV12 mode was used where possible.


Plain avs performance in YV12 mode was 53.03 fps.

Code:
       decomb4-YV12 | ivtc22 BF | ivtc22 FD | FF FD
Film :   16.88*          15.65*     15.61*     49.16*
60fps:   12.21           13.75      11.54      24.57
crap :   11.83           13.92      11.78      25.87
mouth:   14.50*          14.06*     14.05*     28.59

* means deinterlacer (postprocessing for decomb) was disabled.
It seems the perfomance of Decomb depends on amount of frames that need to be deinterlaced.

Film clip was in 16:9 and it seems cropping gave me some speed. It seems speed of decomb mainly depends on the amount of frames that needed to be postprocessed.

FF FD looked best for clip1 and worst for all others. As actual force film doesnt require any processing power, we can see the speed drop from FD, and it is pretty big. I should note that actual deinterlacing takes quite an amount of cpu time, and the detection is pretty speedy.

Decomb clips looked clearly better in FILM parts, due to lack of deinterlacing. Blendfields was a blurry shit, as expected.

60fps parts all looked bad, except that FF did provide a smoother picture. :/

Also, film clip was processed correctly by all tools. Mouth type clips seem to be able to be processed by ivtc22 well if right settings would be chosen. So there, no one questions ivtc22's ability to match fields. It's the need of deinterlacer for lots of sources.


P.S. I remember testing decomb 1.8 vs ivtc21 and decomb was slower about 2 times back then. Great job on optimising it, DG.

Merry Cristmas to all that kept this thread alive (expecially to bobotns)
__________________
___________________MooPolice is watching you!____.o/________

Last edited by OUTPinged_; 25th December 2002 at 19:24.
OUTPinged_ is offline   Reply With Quote