Why was temporal filtering never implemented in Convolution3D YV12 version?

[dosdan]

I was originally delighted about how much quicker it is than the YUV2 version.

But now I realize it only doing spatial filtering. A few days ago I noticed it did not do a particularly good job with a low-light miniDV clip I processed and then I noticed that the last version (Beta 5) still had not implemented temporal filtering.


Oh well, back the the YUV2 version.

[Zarxrax]

Use deen. It has a convolution3d mode that does the same thing.

[Wilbert]

You are wrong, it has temporal filtering. Only the temporal influence (for scene change and fading) is missing.

[Chainmax]

What would be the result of that?

[Boulder]

In other words, scene change detection doesn't work.

[kassandro]

If there wouldn't be a temporal component in Convolution3D, it should have been called Convolution2D.

[Wilbert]

Yeah. So, it means you will/could see ghosting during scene changes.

[Chainmax]

That's no good. Is there a way to translate C3D settings to Deen("c3d",....)? If so, I'd like to know the closest setting to Convolution3D(0,6,12,6,8,2.8,0) (aka preset="animeHQ")