New motion adaptive temporal smoother

[neuron2]

I'm releasing a first beta for my new temporal smoother. I find that it can give excellent temporal smoothing without ghosting. Your feedback will be appreciated. Subsequently, if it is thought worth pursuing, I will describe the theory of operation.

http://shelob.mordor.net/dgraft/DGdnr10b1.zip

The zip file contains help text.

[Koepi]

Damn, I'm in the middle (well, in the first 25%) of a test of my new statsreader tool and thus it'll take some time until i can test that filter (and I'll do then, for sure!).
But it _sounds_ good. The theory behind is still interesting for me, though!

Thanks again for your work,

best regards,
Koepi

[soulfx]

Right now I run temporalsmoother through MAM to keep it from producing weird stuff at fadein/outs and other motion. Would this be close to what your motion temporal smoother does or am I way off?

Whenever I get the chance I'll give it a test. I've just finished running some tests with C3D and it's time for me to get to bed.

Peace,
SoulFX

[OUTPinged_]

Test clip with default settings produces artifacts around edges.

[OUTPinged_]

more comments:

If not taking artifacts into account, the filter runs pretty well. The motion detection part is fast and good quality.

[vlad59]

Hi Donald,

It sounds promising.

The motion detection work well : 50 was not enought to catch all moving object but when I lower the treshold to 25 I can almost see no ghosting (in fact a little one but my sources is very strange).

My test source is very noisy and the filter even with strength 5 don't smooth enought.

Just a funny thing : the filter goes crazy when you go backward in Vdub ;)

Good work.

[neuron2]

Thank you, Gentlemen. There is much scope for improvement, in both the artifacts and the speed. Watch here for further versions.

Yes, I'd already concluded that strength 5 was too low and had increased it to 7. What do you think is a reasonable maximum?

[OUTPinged_]

is it comparable to temporalsoften's luma/chroma thresh in any way?

[drizztcanrender]

Very nice filter,really really fast for a beta version.Although i couldn't see it in terms of quality cause my source was very clean.
First problem i see is that when i set strength to 1,it said strength out of range(2-5).In the readme it says 1-5.
Now for the speed agian what was weird is that i didn't see much difference when changing the threshold number,say from 50 to 20.That's good right?
It looks promising and i'll try it as soon as i get my hands on a noisy source!

[Boulder]

Good work once again, Donald, it's fast and really useful. I don't have any extra noisy source to work on, but I think this filter will become very useful whenever I'll have to encode a movie at a rather low bitrate (less than 1700 or something like that for MPEG-2).

Edit: dammit, I screwed up the speed test between TemporalSmoother and DGdnr.. it seems that DGdnr is a bit slower but as it's an early beta, this will probably change (I hope!) ;)

[Dali Lama]

Hi Donald.

I got this artifact with your filter using these settings:

DGdnr(false,25,5)

The image is attached.

I'll do more testing soon

Bye,

Dali

[Dali Lama]

I tried default settings and similar artifacts appear.

This basically supports Outpinged_ results. Except I believe there are more artifacts in my clip.

Dali

[Boulder]

I also got some artifacts, they appeared at all strengths (2-5). The ones I noticed were in high motion scenes so they could not be seen during normal playback.

[Tri]

Hi Don :)

I have a suggestion: Could you include an option that would exclude colours from being smoothed? E.g. black or very very dark colours so there will be less blocking or smearing during opening credits or space sceenes etc...