PDA

View Full Version : Mpack & focus2


Marc FD
18th August 2002, 16:31
Hi :)
I had a (dumb?) idea in holidays :
an avisynth-toolbox.

several generic filters :
- Convolution/Median/Interpolation (in 1D X,Y or Z axis. Luma/Chroma only modes, optionnal thresholding and clipping, custom matrix)
- Calc filter to do some easy Calculations (add/mul..)

and some mask based filters:
- Mask creators filters (from intensity/delta maps)
- Masks optimisation for above filters.
- ApplyMask filter.
to do conditionnal filtering.

In YUY2. the masks are separate YYYY (fake YUY2) clips.
SIMD optimised in a long-term...

I've already implemented some of these tools to do an easy task like black expand.
but my MMX code is broken :( (C is still working but i'm aiming more speed)
I'll work on it and release a bin soon :D

i would really like feedback on this project. Is it a good or a bad idea ??
(because it's too much work if nobody is going to use it)

(PS : Focus2) I've MMXed Blur/Soften. i've some problems with sharpening but the blur mode works 100% the same (and it's 4.5x faster :)) it could be added to the CVS when it would be finished ?

neuron2
18th August 2002, 17:14
It's not clear what your basic concept is. Are you just packing together a bunch of simple filters that can be used in traditional scripts, or is there something more to it? You said in another thread it would allow users to write filters without coding, but this is not emerging from your description above.

Marc FD
18th August 2002, 19:28
i've uploaded a demo.
take a look :)
(it's a BlackExpand filter)

neuron2
18th August 2002, 19:43
So you are just making some basic filters and packaging them. I was thinking something more along the lines of Avery's filter factory.

Marc FD
18th August 2002, 20:48
?? :confused: could you explain more ???

if you think it's really too crappy, i would simply stop to work on it ....

neuron2
18th August 2002, 21:53
There's nothing wrong with writing basic filters that can be combined to create useful functions!

I was thinking of a high-level filter creation language that insulates the user from the low-level details, would probably be interpreted, etc.

dividee
18th August 2002, 21:58
Hi Marc,
I'm not sure either where you're heading. Seems to be this is very similar to "Layer".
The problem with Layer in YUY2 is that you don't have an alpha channel. So you used this "mask" clip. I used a similar idea a long time ago when I ported Donald's SmartDeinterlacer to avisynth (see MotionMask & MaskedMix). But I didn't grasp the full potential of this back then (it was one of my first plugins), until "Layer" came out.
I'm not sure i understood it correctly, but it seems the "Calc" filter is performing calculations on the mask clip. It seems a better idea to me to expand Layer to work with an additionnal argument in YUY2: the mask clip. But I would use a "regular" YUY2 clip as mask (i.e. keep the chroma channels, but don't use them in Layer). This way you can use regular filters to perform operations on the mask. For instance, "Add" and "Mult" can be performed by the Tweak plugin ("Add" = brightness, "Mult" = contrast).

Marc FD
19th August 2002, 09:39
hmmmm.
I really don't have your knowledge.
Yep, maybe extending others filters would be better.
but to be honest, when i tried Layers & Mask, i was a little lost (BIG LACK OF DOC !!)
Are you sure that using full YUY2 mask is usefull ?? I thinked of this too, but i thinked it would be 2x slower (than YYYY)....
I thinked of YUVA too, but it would be as slow than RGBA

PS : no, i'm not going to create a high-level language.
I'm not as good than avery (and i'm far younger too ;) )