Progressive Refinement

Progressive Refinement:

Basics

• Full matrix solution is slow, with no partial answer
• Requires storage and time, since we need to store form factors for entire calculation
• Recall our equation says how much radiosity is comes from patch j to patch i:

  

  Light is being gathered at patch i.

• Instead we can ask ``how much radiosity'' is shot from patch i to patch j?

  

• Idea: Choose a patch, shoot its radiosity to all other patches, repeat.

Progressive Refinement:

Code 

Keep track of total radiosity (B) and unshot radiosity (dB)

Procedure Shoot(i)
For all j, calculate all Fji
Foreach j
    drad = pj*dBi*Fji*Ai/Aj
    dBj += drad
    Bj += drad
endfor
dBi = 0

Call from

while (unshot > eps) do
    Choose patch i with highest dB
    Shoot(i)
end

Progressive Refinement:

Analysis 

• Shoot lights first
• Can stop and look at partial results
• Can be formulated as a matrix equation
• Lots of cute tricks:
  • Add initial ambient to all patches so that initial results more viewable
  • Can move objects and lights after calculation has started
• Technically, this is no faster than full matrix, but we can stop early
• If we play tricks it is faster than full matrix

Radiosity Issues

  

Questions:

• How big should we make initial patches?
• When do we stop Progressive-Refinement?

Problems

• T-vertices and edges cause light/shadow leakage
• Occlusions and shadows hard to model
• Form factor estimations often cause aliasing

Extensions

• Methods for adding specular (but expensive)
• Substructuring improves quality without increasing big-O
• Smarter substructuring reduces big-O