Fader.dll

[Marc FD]

I am dumb ?? (interessing question :) )
i see don creating a cleaner. (Msmooth to name it)
and it makes me the want to write one myself :D
so here is "Yet another Cleaner for Avisynth"
It's slow & the algo is basic...
but i think it could merit some tests, no ?

works in YUY2. Size MUST be mod 4
i didn't really tested values, so
i would ask the (very few) testers to
tweak it to ill and to try almost all
settings from no-op :
Fader(false,0,0,0,0,0)
to Mega-Average blur :devil: :
Fader(false,256,20,20,20,20)
beware : damn slow !!! (80 iterations...)

[Marc FD]

damn, forgot to attach it :rolleyes:

should be here.... just wait for a mod...

[neuron2]

@MarcFD

I deleted your attachment because the readme file was...well, you know what I'm talking about. If you choose to attach it again after cleaning it up, please also post a short theory of operation explanation so people know what your new filter is all about.

Thank you for your understanding.

[Marc FD]

okay. (thx don)

here is something serious (of course, the readme, not the filter :D )

I forgot to say that this filter is only a sample. dunno if it a good concept.

but when i tested it, i was impressed buy the deblocking.

in fact it only work in wide plain areas (anime ;) )

[Defiler]

Damn, now I'm curious to see the original readme. Heh.

[Marc FD]

ask to don ;)

back to my filter, a little explanation :
It's going to average packs of non-edged blocks, so it destroys some blocking due to quantization in plain areas (where they are _very_ noticable)

i created it to deblock some anime in MPEG-2.
Because many cheaps dvds of anime (let's say 99%)
are more compressed than movies (30 min / GB instead of 15min / GB)
and common MPEG-2 compression artifact (and the one i most dislike) is some blocking, especially noticable in plain areas.

@others anime fan (like me)
right? did you experienced what's written here too ??

[Suikun]

True, I have experienced this, too. Again it's very noticible on TFT displays.
Even in plain blue sky or faces you can see blocks moving around, even without showing the motion vectors. If the filter is really able to correct that I'll be happy to test it throughly >:)
I'm eager to see the sourcecode for it, as I also tried to code such a thing. But my way of thresholding was somehow bad, so for threshold = <anything else than 1 or 0> half of all gradients were gone -_-;;

[bill_baroud]

@Mark FD
i love you =)

since i'm a fansubber, and the raw we find are always _*****_ (or almost) this the most annoying effect i can have ...

i'm trying from a long times to get rid of that, but no filter has do what i'm trying to have ... perhaps it's impossible, but i want to take a look to your filter =)


and perhaps discuss it with you ^^

[Marc FD]

Originally posted by bill_baroud

perhaps it's impossible, but i want to take a look to your filter =)


impossible to recover the whole quality for sure, but if my filter is able to help you recover a little something, i would be pleased to put more efforts in.



and perhaps discuss it with you ^^

no pb, use the thread :)

[Marc FD]

Originally posted by Suikun
True, I have experienced this, too. Again it's very noticible on TFT displays.
Even in plain blue sky or faces you can see blocks moving around, even without showing the motion vectors. If the filter is really able to correct that I'll be happy to test it throughly >:)
I'm eager to see the sourcecode for it, as I also tried to code such a thing. But my way of thresholding was somehow bad, so for threshold = <anything else than 1 or 0> half of all gradients were gone -_-;;

i don't think it would make miracles.
(i don't think any filter can make miracles :D )
my approach could be too "destructive".

but, anyways testing and is much better than talking !

and i didn't say i couldn't improve it.
it's very low-level code compatible

of course my source code would be GPLed if it's used (that's why i write it)
but i'll wait for the XviD issue to finish _good_ to do it.
if not, byebye opensource devellopement :(

[Bulletproof]

Not to sound unencouraging, but can't current temporal smoothers already try to fix quantization artifacts?

[Marc FD]

only if they move. and if they move, temp filters would ghost....
but, yes, temporal filtering helps too.

(and it's so easy to do ;) )