author of this piece speaks strong words against donald graft's kerneldeinterlacer...

...as i suggested some things to donald(regarding inclusion of motion-map deinterlacing) and he included those into KD(after that kd became THE deinterlacer for avs) i felt a bit insulted myself...so i said;what the heck;he who says KD is crap,must of have made better deinterlacer than graft did......i'll take a quick look...

so i did...but he didn't....

http://i4004i4004.bizhosting.com/alignfields/

his doc(too big and confusing as usual...removedirt style) is also full of wrong info (for example it is stated the dgbob is the first framerate doubler...yeah,right...)
and he concentrated all the effort (i read this between the lines of the doc)in beating the (essentially) lousy deinterlacer;TMC....

i already said that if morsa post images that look lousy then deinterlacer looks lousy
http://www.hydrogenaudio.org/forums/index....ndpost&p=194126
(on the end of this post)

also in the images i posted experienced eye will notice that this is a "dumb" deinterlacer....it is deinterlacing still parts too(notice the subtitles and tv-station logo ("HRT1")...see some parts of it missing?) and that was the "best" mode (mode=0).....

sorry kassandro,but you messed this time;removedirt is nice vhs cleaning filter,but this is

doom9 thread here:
http://forum.doom9.org/showthread.php?thre...20&pagenumber=2

edit;typos,smilie,doom9 link....

hehe. harsh review indeed... but pretty accurate atm.

it's a VERY noise-sensitive deinterlacer... hence the _almost_ dumb bob you get out of it.

the concepts discussed are of course valid (and exhaustive!), so perhaps updates of this will prove more useful.

i feel it's best not to dis any new filters but rather offer constructive criticism (especially for smart bobbers - these are more than welcome in my opinion, good or bad), and AFAIK Kassandro is aware of the issues in alignfields.

Kassandro appears to be harsh on fellow devs, but he's proven himself a capable programmer...

i will watch alignfields' development closely

[edit]

an easy way to prove it's at least attempting a smart bob is to feed it a duplicate, interlaced frame. you'll notice that it comes out untouched, due to it being recognised as being not combing, but detail that persists over 1 frame. of course we know it's combed, but if computers were as smart as us...

i don't think bobbing is a good idea(generally i don't watch video on monitor but via tv-out) so i prefer to find a better way to do interlaced encoding,than to do bob....

and i just found a way to do interlaced mpeg2 that looks as good as interlaced mpeg4...mencoder....and it'll play on my tv-out just fine.... even the 720x576 streams...SWEET!

but if you must watch it on monitor,then you must bob.....if scharfi says he's ok with his bobber,then i believe that works fine...

kassandro is ok,but i won't let anyone make false claims...(although siwalters,morsa and sharfi were all faster than me (to reply)...hehe)

tell him to make FAST motion compensated denoising;that's what we're waiting for...(wink)

i agree there, however smart-bobbing isn't really for display (indeed a good enough dumb bobber will fool my eyes into seeing detail and motion on a progressive screen). the real strength of a good smart bobber is for "difficult" telecines, or worse, NTSC->PAL converted content. we want to get rid of blends, but not lose detail. the only way to do that ATM is smart bob -> deblending algo -> decimate/re-interlace.

scharfis' bobmatcher works very well, but there's still artefacts getting through with certain content (and difficult to predict when this will happen)... indeed my experience with smart bobbing is choosing one that gives you the least distracting artefacts, trading off smimmer, blur, and resident combs/blocks/other weirdnesses.

personally i use a home brewed script, but this blurs the image a tad and sacrifices a bit of chroma res as well. (on converterboxed content it's good, as usually these poor sources really don't have much left to lose)

man, i'd love to have a really good TV-out! (and a TV in, for that matter... but in my region it's a bit wasted, as i'm too far away from the broadcast towers and don't have a big fekoff antenna... doesn't matter. nothing good on teev anyway)

[edit]

definitely going to have to poke around with mencoder... i'm always on the lookout for a good mpeg-2 encoder. mpeg-4 players haven't reached critical mass in Oz, and besides that i can't afford one.

i don't think there's a good solution to make 25fps blended content into 25fps ok content.....(if blending pattern was irregular...sharfi did a nice job on deblending regular patterns....)
is that what you were talking about?

hehe, I know my bobmatcher has stairstepping artifacts on slow moving borders/lines
and remaining combs on chroma-only moving areas.

everything on my bobmatcher stands and falls with the comb-mask.
the comb-mask of mpeg2dec.dll is okay, but does not catch >1 Pixel-movements, has too much post-enlarged-deinterlacing-areas (sometimes blockiness), and does not have a chroma-test.

my test extracting a map of swdgbob.dll were sucessful. but its map sadky did not fit to any video. It left combs everytime....

Has anyone a good bobbing-comb-mask-algo to offer?


@ivo: with non constant pattern go for restore24. I've updated it a little bit. It works very well now. especially on music-clips.

nice to hear it....but..have you tried ,mencoder mpeg2?(hehe)
VERY fast too...
anyway,progressive playback is much easier to setup (regarding my tv-out),so i'll try your stuff.....
restore24 you said...

edit:or will i...i think this would drag at about 0.0fps on my machine....
anyway,thanks for the tip.....

i must add something!
kassandro said:

in
http://forum.doom9.org/showthread.php?threadid=76458

boy..ohh..boy...for sure he didn't used his eyes when he made this filter!!!


btw.graft getting real angry lately....people BE CAREFULL!!!