can't activate mode 2 for temporalsoften

[theReal]

I was reading about the new mode 2 for temporalsoften() in avisynth 2.5 - I'm using avisynth alpha 01/10/03 and everything is working perfectly so far, but temporalsoften still only accepts 3 arguments.

Shouldn't I be able to activate mode 2 like this:

temporalsoften(4,8,4,2,5) ??

Because when I do so, avisynth says "invalid argument for temporalsoften". The only way I get it to work is the old way with 3 arguments (like "temporalsoften(4,8,4)")

I thought "[int_mode,]" stands for "integer value that sets the mode, optional argument"
Am I wrong with this?

btw. I'm using an Athlon XP Palomino and WinXP Pro, SSE is enabled.
Or does mode 2 require SSE2?

[sh0dan]

Download new binary (gives better quality), try: temporalsoften(3,10,10,scenechange=10,mode=2)

Mode 2 now also now works on all computers. Scenechange still requires K7, P3 or newer.

[theReal]

thanks - I just saw the changelog that said the new temporalsoften mode 2 was introduced in the binary from the 17. of January 2003 - on avisynth.org it said "Starting from v2.5 two options are added to temporalsoften". That's why I thought it should work with the alpha from the 10. of January as well.

My problem with the new version (18.01.2003) is that it says "will not load non-updated plugins" and I couldn't find any updated versions of crucial plugins like decomb, convolution3d and warpsharp (I'm trying to use 2.5 alpha for all my encodings, not just for pure testing, so I really need those filters).

[sh0dan]

decomb is updated, warpsharp (at least the one on the alpha site) is updated. Until the new convolution3d is uploaded you can get it from this url (http://cultact-server.novi.dk/kpo/avisynth/Convolution3D.dll). Don't spread the link - it'll be removed shortly!

[Wilbert]

Download new binary (gives better quality), try: temporalsoften(3,10,10,scenechange=10,mode=2)
Oops, I switched the last two arguments on avisynth.org. I will correct it soon.

[theReal]

Thanks! I'll get the new binary then.

[Guest]

Originally posted by theReal
I couldn't find any updated versions of crucial plugins like decombDid you try the author's web site, not to mention the thread "Decomb beta thread"? :)

[vlad59]

Originally posted by theReal
thanks - I just saw the changelog that said the new temporalsoften mode 2 was introduced in the binary from the 17. of January 2003 - on avisynth.org it said "Starting from v2.5 two options are added to temporalsoften". That's why I thought it should work with the alpha from the 10. of January as well.

My problem with the new version (18.01.2003) is that it says "will not load non-updated plugins" and I couldn't find any updated versions of crucial plugins like decomb, convolution3d and warpsharp (I'm trying to use 2.5 alpha for all my encodings, not just for pure testing, so I really need those filters).

C3D is now officially updated :
http://forum.doom9.org/showthread.php?s=&threadid=38281&pagenumber=3

I'll contact Defiler to update Web page.

[theReal]

Did you try the author's web site, not to mention the thread "Decomb beta thread"? Oooops, sorry neuron2 - I guess I should have found the newest decomb beta because the link from avisynth.org goes directly to the decomb beta thread you mentioned (it was mainly convolution3d that I couldn't find).

I guess my not finding this and that appears pretty stupid to someone who is involved in the development of avisynth 2.5, but it's a little harder for me to pick the right versions of everything together from all the different sites.

[sh0dan]

I must say, the new mode and scenechange detection really helps temporalsoften - even more than I expected. My results show that it works very well as a very fast alternative to convolution3d. Performance is far above realtime on DVD-size images - nice :)

[WarpEnterprises]

what exactly does mode2?

[sh0dan]

In the old-fashioned temporalsoften, pixels are calculated like this:

new_pixel = (pixels_within_threshold) / (number_of_pixels_within_threshold)

So the new average pixel is calculated from the number of pixels within the threshold. Mode 2 does:

if (pixel_is_within_threshold) {
new_pixel += pixel_within_threshold
} else {
new_pixel += original_pixel
}
new_pixel /= kernel; (kernel is the number of possible frames).

That means, that if a pixel in a frame is not within the threshold, the original pixel (from the current frame) is inserted instead.

oh - and btw - for those interested - scenechange is not calculated between adjacent frames, but from the current frame. That means that all frames within the kernel are tested against the current frame, and not the frame next to it. This improves fades and other gradual transitions, and kills almost all residual ghosting.

[vlad59]

That's the same trick I use for Convolution3D (to always divide by the same value), I should have a look to your code and see if I made crappy optimizations.

[sh0dan]

I know - that's where I got the inspiration :)
They are very different. I'm still working on an optimized c3d, but I can't seem to be able to squeeze much more out of it - but I may be able to make it better quality without any penalties. Right now I'm doing a full matrix operation on all pixels at roughly the same speed as ordinary c3d, but I'd like to do a temporal-disable function, just as ordinary c3d.

Still trying some final attempts - it might turn into a general-smoother function - but it might not turn into anything usefull.