BiDir: Distance dependent brightness on certain light paths

[CodeFHD]

Attached is a Blender test scene with a simple setup designed specifically to test some aspects of the BiDir engine:

• Two square planes with Matte material, at 90° angle to each other with a bit of space in between
• Two light sources: A Laser and a Spot with narrow angle. The light sources point at the center of the first plane
• A camera, pointing at the center of the second plane (not actually critical where it points exactly)
• Three thin Glass blocks: One in front of the light sources, one in front of the camera, one between the two planes.

bidir_distancefalloff_V3.zip

The purpose of the glass blocks is to force certain connections to be possible or not, as a direct connection can not be made through a specular material:

• The one in front of the light source blocks Direct light sampling
• The one in front of the camera blocks direct camera connection
• The one between the planes blocks direct connection between the two diffuse surfaces. In this simple setup, effective allowing only pure path- or light-tracing

The setup works together with very low path depth settings: 1 for eye and light each by default. When activating a glass block, increase the path depth of either light or eye to 3 to allow specific paths.

From this setup, the following observations are made

• In the default setup, moving either the laser light or the camera along its local z-axis (i.e. closer or further away) changes the rendered brightness of the image.
• The above effect does not happen with the spot light, indicating a specific issue with the laser light
• However, adding glass increases the brightness also for the spot light. Applies to light path depth >1, too.
• When setting the light path depth to >=2, the camera seem to retain brightness.
• The laser light falloff happens regardless of path depth settings
• Glass in front of camera or laser fixes brightness falloff for that object (but not the other)
• With glass between the planes and light depth >=4, the camera falloff is fixed
• With glass between the planes, light depth = 1 and eye depth >=4, the second plane remains dark.
• For those setups where the brightness depends on camera distance, the brightness also depends on the FoV size. Can be triggered either by changing the camera focal length, or setting a render subregion in the viewport.
• When switching to orthographic camera, the 1-1 path depth case remains entirely black

Short summary list of issues to investigate and fix:

• Laser light distance falloff
• Brightness increase through glass for spot light
• Camera distance falloff
• Why Eye-path route stays dark with Center Glass
• Brightness scaling with FoV
• Black image with Orthographic camera

Initial ideas for investigation

• MIS factors seem to affect the behavior

  LuxCore/src/slg/engines/bidircpu/bidircputhread.cpp

  Line 887 in cb7e64d

  eyeVertex.dVCM *= MIS(eyeRayHit.t * eyeRayHit.t) * factor;

• Handling of cases around variable sampleResult.firstPathVertex
• f-tan-theta processing of perspective camera involving FoV and screenWindow variables

  LuxCore/src/slg/cameras/perspective.cpp

  Line 80 in cb7e64d

  void PerspectiveCamera::InitCameraData() {