interior test [Distributed Path integrator explained]

[Ukko]

I have a scene where the lamps on the ceiling and walss have lampshades (cloches). I cannot handle that scene without mattetranslucent or roughglass material. At now i am experimenting with mattetranslucent material and increasing diffuse reflect and refract samples to gain a clean image, but still cannot find any solution. The material behaves like something almost without transluency - only few noisy spots appear in the nearest area around the cloche, rest of the scene id dim. Maybe i'm going a wrong way? Maybe someone knows how to make mattetranslucent work properly with distributed path?

[jeanphi]

Hi,

The number of samples is probably irrelevant at this stage (ensure you have at least 1, that will be enough), what is more important is the depth: if you set it too low, you won't be able to reach the light source, set it to at least 3 or 4, maybe more and then go downward until you find the difference is unacceptable.

Jeanphi

[binarycortex]

I have a scene where the lamps on the ceiling and walss have lampshades (cloches). I cannot handle that scene without mattetranslucent or roughglass material. At now i am experimenting with mattetranslucent material and increasing diffuse reflect and refract samples to gain a clean image, but still cannot find any solution. The material behaves like something almost without transluency - only few noisy spots appear in the nearest area around the cloche, rest of the scene id dim. Maybe i'm going a wrong way? Maybe someone knows how to make mattetranslucent work properly with distributed path?

Exphotonmap may be a better fit for this scene if DP is having issues. It's easier to tune and gives incredible results now that it is fixed. Also good old metropolis + bidirectional is still the best at finding difficult light paths.

[Ukko]

MLT + bidir is still the best, but still the slowest. I resigned translucent cloches, just apllied light material to them (instead of making a light bulb inside translucent case). But another issue appeared. A problem with mirror. There are areas that should be iluminated (so their mirror reflection should be too), but they are dark regardless of how many bounces I set for glossy, diffuse and specular. The scenario is simple - ligh comes from the meshlight on the ceiling, hits the mirror, hits the part of the wall (which mirror reflection is marked), hits the mirror again and goes through the camera, so recursion depth set to 4 should be enough. Why it is not?

EDIT: I solved it with Photoshop, just copied a piece of image and pasted in the problematic black area. Below is a link to an image how this scene should look like, but not always it will be that easy. http://greenpencil.eu/images/wnetrza/lazienka.jpg

[binarycortex]

MLT + bidir is still the best, but still the slowest. I resigned translucent cloches, just apllied light material to them (instead of making a light bulb inside translucent case). But another issue appeared. A problem with mirror. There are areas that should be iluminated (so their mirror reflection should be too), but they are dark regardless of how many bounces I set for glossy, diffuse and specular. The scenario is simple - ligh comes from the meshlight on the ceiling, hits the mirror, hits the part of the wall (which mirror reflection is marked), hits the mirror again and goes through the camera, so recursion depth set to 4 should be enough. Why it is not?

issue.jpg

EDIT: I solved it with Photoshop, just copied a piece of image and pasted in the problematic black area. Below is a link to an image how this scene should look like, but not always it will be that easy. http://greenpencil.eu/images/wnetrza/lazienka.jpg

I would like to see this scene done with exphotonmap, just to see what it does.

[jeanphi]

Hi,

Are you sure of your materials and geometry because just below or above the black area, it's not black.

Jeanphi

[Ukko]

Jeanphi - I'm sure about the normals (I've checked again) all they are pointing in the proper direction. The material is a bright gray glossy with low exponent and specular value (exp=5, spec=0,01). The areas, which You've mentioned get light directly from the cloche, or reflected by walls. The marked area gets light from the mirror, then reflects it, the light hits the mirror once again, before it gets to the camera. Maybe thats the reason (just my hypothesis).

binarycortex - when I find some time, I'll render this scene with photon map and post the result.

If you know about other user's tests with DP and mirror, please post a link to the proper discussion.
Thanks in advance !
ukko

[jeanphi]

Hi,

I still get the impression that the black area is receiving light directly from the spots in front of the mirror. I'm almost sure that the 2 top vertices of this face have their normals reversed. Look at the smooth transition to black at the bottom of the red box.

Jeanphi

[SATtva]

Ukko, make sure the face isn't smoothed.

[Ukko]

Finally I found some time to make some experiments. Everything was fine with the mesh - there wasn't any inverted normals, or smoothed faces. What did show up, was the problem with the Rejection system. It seems not to work properly with mirror material.

rejection_settings.png (4.26 KiB) Viewed 439 times

I made few border-renders for the problematic area, and the results are below.

lazienka_dp_rejection.png (8.26 KiB) Viewed 439 times

lazienka_dp_no_rejection.png (10.44 KiB) Viewed 439 times

lazienka_expm.png (10.32 KiB) Viewed 439 times

The first one from the left is made with DP and rejection on, second is made with rejection off, third is made with ex-photonmap.

Of course there are lots of fireflies at the same number of Spx, but the main issue is solved. I've tried to render the scene with rejection on, and the treshold at the highest possible value (10), but the result is the similiar to the first image - black wall. My next step is to change the mirror to a high-exponent glossy material.

Windows XP, LuxRender 0.8 RC4, Exporter v. 0.7 for Blender 2.49