I'm learning light probes used for dynamic global illumination. I have a question regarding the placement of light probes, as based on most of the pictures I have seen, they seem to be placed uniformly in a grid. Is that a reasonable way for placement? I feel that more should be placed in corners than in the middle of an area.

is there any research on light probe placement that minimize the overall data needed for rendering?

this GDC talk http://twvideo01.ubm-us.net/o1/vault/gdc2012/slides/Programming%20Track/Cupisz_Robert_Light_Probe_Interpolation.pdf

mentioned irregular placement on tetrahedrons and how to do interpolation. but it doesn't seem to say much about placement itself.

this paper http://melancholytree.com/thesis.pdf mentioned that only one probe is needed in a convex shape, but I don't seem to see people do this, is it because this is only for static global illumination without moving objects?

what's the the latest development in light probe placement?


Looking for answer for similar question, I found that work: http://www.cse.chalmers.se/~uffe/xjobb/bowald_final_master_thesis_v2.pdf

It is based on automatic grid placement which depends on scene voxelization. Inspired by this I developed similar technique:

  • First scene is voxelized with small voxel size (binary voxelization for memory saving)
  • Result is downsampled to desired grid size (size of each cube with eight probes on each corner)
  • Based on result - data structures are prepared
  • Empty areas are optionally computed - cubes size can be bigger to save memory of volumes and probes buffers but it must be always multiplication of smallest cube size to fit in the grid
  • Lighting for each probe is precomputed

Data structures are:

  • Volume map buffer - mapping world position to volume index
  • Volumes buffer - each is array of eight probe indices
  • Probes buffer - SH data of each probe

This way performance is constant, fast and independent of quality (size of cube) unlike in tetrahedra grid searching algorithm.

However I don't know if there is more recent advancement in that field. You might check http://jcgt.org if there are any new research papers.

| improve this answer | |

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.