LuxCoreRender Forums

I have been trying to do more modelling of lens systems, this time with a simple Cooke Triplet. blend-file attached.

In my scene I only have a 5 cubes with emission (they produce some defocus circles), and a Suzanne model illuminated by a Sun lamp. With these being intersectable sources, rendering works well with both BiDir and classical Path tracing:
However, if I add light tracing, I get the following output:
There is some indication of this in BiDir as well if you crank up the gain (also the cubes are quite small, so Metropolis has some issues and didn't find one cube in this case, but I guess that is to be expected)
Any mistake on my side, or is this a bug?

CodeHD wrote: ↑Tue Jul 09, 2019 1:47 pm With these being intersectable sources, rendering works well with both BiDir and classical Path tracing:

BiDir and light tracing are useless to render your scene because light path vertices on specular surfaces (i.e. lens) are not connected to the eye. So all the not black samples are coming from eye paths (i.e. normal path tracing).

Also calling the sun light intersectable, is only a theoretical exercise: it is easier to win at the national lottery. Try to increase the sun size (relsize parameter) by 10 or more and the intersection will start to be a bit more likely.

Also the overall scene size is likely to suffer of numerical precision problems with so huge distances compared to object sizes.

Ok I relaize I had some wrong thoughts earlier, plus some info missing:

- I thought of the Suzanne head as intersectable, but it is not an emitter itself. However, it is matte, so you can connect to the Sun from there.
-Also, there is a "detector" behind the lens, the usual matte-translucent slab. So there is a diffuse surface there as well to connect to the camera. (sam results btw. if you make it a simple matte material and observe with the camera from the front)

Numerical precision could be it, increasing min-epsilon seems to make the problem go away...