I also sent this post at odforum
Getting a bit confused/angry with mantra renders.

Hi, i'm having some rendering problems…

I checked my topology.
There is still rendering artifacts



Here is the displacement shader:
each normal of the “canyon walls” is smoothly interpolated, in the displacement shader, between the geometric normal and the “up normal”, with the height, so that it creates a correct transition between the walls and the plateau.
(normals are smoothly interpolated between each face of the walls but not between walls faces and plateau faces)

I tried to adjust :
-the displacement bound param. in OBJ parameter page
-the shading Quality param. in OBJ parameter page

-the Super Sample param. in MantraROP parameter page.

But still these stupid artifacts.


Where can they come from?

Division by zero?
Still uncorrect geometry?
Bugs?
insufficient displacement bound? -> increased it and either no change or never-ending renders.

Any idea or help is welcome!

Are you using subdivision surfaces? Have you tried debugging the shader i.e. remove the stuff you're doing with the divide/smooth/mix and simply apply a displacement to the whole thing? Does it go away? Still there? This helps to isolate whether the problem's in your data or your shader. Simplification and isolation are your friend…add elements slowly and see where it breaks.

Also, upping shading quality probably won't help you, and if you make some changes to displace bounds with no apparent change to the result, that isn't your problem…

Cheers,

J.C.

simplification and isolation…
The painstaking and so necessary methodology!
Mais ou avais-je la tête!
i thought ( and wished) there would be a faster answer.

I found it at last:
it comes from specular computations within my shader. I turned to a Lambertian shader…
Noise+vexSpecular seems to give ugly results

There's no simpler way than debugging and simplifying. What you might save in time only increases wasted time on this end of people trying to guess all the variables you may have, which in turn wastes your time all over again.

Cheers,

J.C.

Its pretty easy to get artifacts with the specular functions. If I'm not mistaken and can remember correcty, the specular functions return NaN if one of the supplied vectors are zero length.

Thanks,
Jason