[Micha]

Hallo,

I have render a scene with only one dielectric object and some plastic/diffuse/metal. After 4 hours I have got a nearly noisefree image, but only the dielectric is full of noise. I attach an image (sorry, I can not show all). My question: What can I do?

I think, if I let it render longer, than the full image will be sampled again and again ... but only the small dielectric object need more samples. My suggestion: It could be nice to get a parameter to sample some materials more than other materials.

[Maximus3D]

Ehm.. not to sound stupid or anything but what's up with all the white fields in the image ? looks like a puzzle with alot of missing pieces.

/ Max

[Micha]

Oh, I don't want to show the hole object, because it is a current project, so I delete some parts. I only want to show, the image is noisefree and only the dielectric is noisy. Now, I'm not so happy to render "three" days only for the small dielectric part.

[Mihai]

Maximus:

I attach an image (sorry, I can not show all)

The sampling per material has been requested a few times and it would be TREMENDOUS, INCREDIBLE, UNBELIEVABLE, if it is possible.

But I don't think so......

Is it a mousetrap? A stapler?



Bicycle lock?

[Thomas An.]

Ehm.. not to sound stupid or anything but what's up with all the white fields in the image ? looks like a puzzle with alot of missing pieces.

/ Max

He is beta testing a pre-release Maxwell patch.

Actually he said, he can't show the whole thing.

Micha: Your idea makes sense and it has been asked before in the wish section (but I can't find the thread right now).

[Micha]

Thank you Thomas without link also.
So, I belive this is a standard problem and NL will try to find a solution.

[Micha]

Psst, it is a secret.

Best would be, if Maxwell would identify noisy areas automatic by itself and use more samples there - adaptive sampling without extra user options.

[Micha]

Hi WolfKiller,

hmm, the normals of the red housing was wrong, all other normals are right and the emitter are single poly meshes. But now, with normals outside it looks like befor.

I have an idea, somewhere I have seen a beta test with a ring and that the reflection of the caustics are noisy. Maybe, this is the same effect here.

Also I could be I have a worse case here. The transparent plastic is a little blurry (small roughness), the red plastic is blurry and the lighting on the red housing through the transparent gear is only caustic.

[Thomas An.]

Micha,

I suggest you wait a little until the beta patches are ready from NextLimit.
Somehow reflected or refracted caustics (as you mentioned) are not as smooth in this beta as it used to be.

[Micha]

In the image above I have used a pysical env without sun (as reflection env) and two emitters. An other new test without rough surfaces show the same noise.

Now, I have done a test with a simplier scene. Here the images done without indirect caustic layers. The red part is plastic.
The dielectric is a problem in all environments, it need to much time compared to the rest of the image

[Micha]

Hmm, the dielectric material seems to be the Achilles jerk. The image is ready, only the dielectric material need much time and the whole image will be sampled more than needed.

[Micha]

Luca, I use a custom material and have set the abbe at 200. I think, I will test the low polysphere and a default material next night.

[Micha]

Sorry, I will make this test later.

But, now I have think about this situation and understand the problem a little bit more. The problem is, the light travel two times through the dielectric material - one time from the light, one time to the camera. This bring much noise. Seems to be a horror situation.
For a simple wine glass scene it is only one time. Here I have seen the same problem in architecture.

http://www.maxwellrender.com/forum/viewtopic.php?t=3984

[Kabe]

Hmm, the dielectric material seems to be the Achilles jerk. The image is ready, only the dielectric material need much time and the whole image will be sampled more than needed.

Maybe NL could implement a mechanism to add an option to oversample dieelectrics - or any other specific material - a bit more? This will still create problems though if a dielectric is reflected in a silver mirror...

Kabe

[Micha]

@Luca: the default material bring some color in the noise only, the custom material with abbe 200 seems to be better. Also the emitter dome dosn't help. Very strong noise behind the dielectric object after 12 hours.