June 19th June 23rd
CDF sample
LightTree sample
- Fixed the tree middle partitioner to produce a balanced tree
LightTree with balanced MiddlePartitioner
LightTree converging
- Added the `inverse_distance_falloff again into the probability calculation
probability = m_node_energy * inverse_distance_falloff * inverse_surface_area;
Converging
probability = m_node_energy * inverse_distance_falloff
Converging
CDF at 16 samples per pixel and diffuse bounce set to 0
Comparison of different partitioners at 16 samples per pixel and diffuse bounce set to 0
MiddlePartitioner
MeanPartitioner
SAHPartitioner
- Fixed a bug when calculating node energy (only the red component was used)
- Compared use of luminance for node energy value to luminance approximation (spectrum average)
- luminance produces slightly less noise in blue area (we can ignore it for now)
Luminance
Luminance approximation
- Fixed distance falloff bug - distance between the sampling point and the bbox was squared twice
1 sppx
Converging
CDF reference
1sppx
Converging
1sppx
16sppx
Converging
CDF reference
1sppx
16sppx
Converging
-
Created unit tests for
MiddlePartitioner- in 2d with ASCII graphics
-
Tested for more than 2 lights
fireflies occur due to probability being evaluated only based on the distance of the light- sampling artifacts appear for samples in between two lights of the same intensity
-
Nathan produced test images with probability being evaluated only based on the distance of the light - it doesn't produce a bug
-
Visualized LightTree structure using VPython
- found a minor bug when defining NLS volume box (AABB constructor arguments min and max were switched)
-
Determined that tree structure looks good and probability calculations seem to be calculated as expected.