Petr Kellnhofer^{1} | Tobias Ritschel^{1,2} | Karol Myszkowski^{1} | Elmar Eisemann^{3} | Hans-Peter Seidel^{1} |
^{1}MPI Informatik | ^{2}Saarland University | ^{3}Delft University of Technology |
Here we provide lookup tables that we used to sample probability distributions on the GPU. We loaded provided table values as textures and used texture sampling in OpenGL shader to read them in real time.
Each inverse cumulative distribution function (ICDF) contains matrix of values separated by spaces and stored as a text file. First row contains number of rows and columns. Following lines each contain one row of a matrix. Each row corresponds to one ICDF of distribution with certain parameters (see lower). The columns uniformly sample percentiles of respective CDF and should be accessed using uniform random variable to achieve sampling of original probability distribution function (pdf). Each value is a value for given percentile of CDF.
Inverse function for F(Θ, μ) from the paper. Used to sample Quantization noise (Eq. 1 in the paper) and Photon-like noise (Eq. 2). Each row corresponds to single value of parameter λ of Poisson distribution starting from 1 and sequentially going to 1023.
» DownloadInverse function for CDF of probability distribution function P_{All}(k, N) from Eq. 3 in the paper. Each row corresponds to single value of parameter N starting from 1 and sequentially going to 1023.
» DownloadInverse function for CDF binomial distribution from P_{Final}(M,N') in the paper. It assumes that there are 4.8 rods per pixel, therefore p = 0.0096. This matches our 22'' 1680x1050 display with 0.254 mm pixels observed from 60 cm and it should approximate common PC setups well. Each row corresponds to single value of parameter M starting from 1 and sequentially going to 1023.
» Download