Light Map Gen

Jeremy Pronk 2005
www.happiestdays.com

Version 1.0beta4

Whats New

1.0beta4

Increased precision in light map files
Outputs sample solid angle (spherical area) to the light file map (current only implemented for Median Cut)
New normalize technique, set total light map intensity to 1 (old technique is called -normalize2)
Runtime speed increased due to new compiler optimizations (Win32 version)
Linux version (x86)

1.0beta3

Added median cut sampling. Based on ideas in the sketch "A Median Cut Algorithm for Light Probe Sampling" by Paul Debevec.
Added "resize" parameter.
Added lightMapGen_loader MEL script

1.0beta2

Initial release.

Overview

lightMapGen is a small application designed to find the most appropriate sample locations in an intensity map.

See quick start.

For specific details on the techniques employed it is essential one reads the following papers

A Median Cut Algorithm for Light Probe Sampling by Paul Debevec
Efficient illumination by high dynamic range images by Thomas Kollig and Alexander Keller.
Interleaved Sampling by Alexander Keller and Wolfgang Heidrich

Usage

The command line program "lightMapGen" accepts a single high dynamic range floating-point TIFF image and a number of parameters. The common use of lightMapGen is to generate a light map from a latlong HDR Tiff image which can then be used directly by sphericalLight to illuminate geometry in mental ray.

lightMapGen may also be useful for those interested in utilizing the light map file in their own light shader. Hopefully the file format is self explanatory.

See quick start.

Light Map Gen Loader

lightMapGen_loader is a simple Maya MEL script which will load a lightMapGen light map as a "dome" of lights. There are simple controls for all the lights in the transform of the parent locator.

What is Interleaving Sampling?

Interleaved sampling is an idea which was central to the creation of lightMapGen. Please see Interleaving Sampling, Alexander Keller and Wolfgang Heidrich for a thorough explanation.

The basic idea is to render with the same number of light samples but to choose alternating (light) sample locations actually calculated from the map itself. This provides for much improved anti-aliasing particularly in the shadowed (or occluded) parts of the render.

Parameters

Required Arguments

Input map (-in filename)
Input intensity map. Must be a valid floating-point TIFF file.

Sample number (-s N M)
N is the number of samples to calculate
M is only used for sampling type 3 & 4 and is the number of interleave levels for each sample. If M is specified interleaved sample generation is invoked.

Optional Arguments

Sampling type (-st sampling-type)
0 - voronoi tessellation (used for debugging)
1 - centroidal voronoi tessellation (CVT), similar to technique used in lightGen
2 - importance voronoi tessellation, much improved variation on CVT
3 - interleaved importance voronoi tessellation, as 2 with multiple interleave levels for each sample
4 - median cut (default)

Map type (-mt map-type)
Format of input intensity map.
0 - flat map
1 - lat-long or spherical map, must have an aspect ratio of 2:1 (default)
2 - test map, for debugging

Resize (-resize amount)
Resizes the input image amount times before processing. Good for up sizing small maps when using median cut and for downsizing large maps we using voronoi style sampling. (defaults to 1.0)

Normalize (-normalize)
When specified, the samples will be scaled so the total intensity is 1 (i.e. all samples add up to 1)

Normalize2 (-normalize2)
When specified, the samples will be scaled so the maximum sample value is 1.

Write diagnostic images (-writeDiagImages)
Writes a TIFF representation of the voronoi tessellation after each iteration. The results is slightly different depending on which sampling technique is employed but general each region has a unique grey level, whilst mass centroids and region generators (samples) are represented by red and green pixels. The output images are floating point TIFFs. See example.

Write density (-writeDensity)
Writes the intensity map of the input image. I.e. the intensity value(s) used internally in the calculations.

Voronoi Specific Arguments

Max centroidal offset (-mo offset)
Maximum offset (in pixels) allowed for an CVT calculate. Note that the max centroidal iterations param will override this. Defaults to 2.

Max centroidal iterations (-mi iterations)
Maximum number of centroidal iterations allowed. Regardless of the current mass centroid offset, there will be no more than this number of iterations. Defaults to 8.

Interleaved iterations (-ii iterations)
Number of interleaved iterations to perform. Defaults to 3.

Radius of perturbation (-rp offset)
When inserting new samples, they will be a maximum of offset pixels from the insertion location. The idea is to keep new samples close to but not on top of the high intensity region generators. The offset is a percentage of image size. Defaults to 0.015 (1.5%).

Tips and Tricks

I recommend using the median cut sampling algorithm, it is very quick and quite efficient.
When using median cut I suggest you use full size image maps as input or use the resize option.

Known Problems

all flavours of voronoi tesselations can fail at higher sampling counts and this is dependant on the input image