Atrous Filter

[namekuseijin]

have the devs seen this already?

http://www.uni-ulm.de/in/mi/graphics/atrous-filter.html

should look great for LuxGPU...

[rafal]

Hi,

I'm just wondering whether the devs managed to overlook this post or so? The filtering algorithm looks so stunning, that if it's not fake, it's kind of revolution . Would it be possible to adopt it for LuxRender somehow? What's your opinion?

[silence]

Don't know why but even if it is impressive, it also seems blurry. I really would like to see a HD version because the low resolution may be the cause.

[jeanphi]

Hi,

Very interesting paper. It requires some buffers that we aren't computing right now, but it might be worth looking into this in the future.

Jeanphi

[B.Y.O.B.]

Don't know why but even if it is impressive, it also seems blurry.

Especially the reflections on the metal sphere in the cornell box look blurry to me. And without noise the image looks biased in a way
Very interesting...

[thomas]

To me it looks similar to the result obtained when using the bilateral blur node in Blender: http://www.blender.org/uploads/pics/biblurnodes.png . Which indeeds can operate on additional buffers such as color, normal and z-depth. I use it a lot.

[jeanphi]

Hi,

As described in the paper, it uses similar tricks to detect edges, however it uses wavelet filtering instead of gaussian blur and is much more efficient than traditional bilateral filtering as is demonstratede by some figures in the paper.

Jeanphi

[B.Y.O.B.]

it uses wavelet filtering

I once tried to de-noise a luxrender-image with wavelet filtering in GIMP - it was awful, destroid all the fine structures and was still noisy (but maybe I didn't fiddle the settings to the very best).