max-uho / fast_gpu_voronoi

GPU-Accelerated Jump Flooding Algorithm for Voronoi Diagram in log*(n)

A Fork of https://github.com/maciejczyzewski/fast_gpu_voronoi for my modern Mac M1

Demo (Google Colab)

	Research	Authors
[slides]	GPU-Accelerated Jump Flooding Algorithm for Voronoi Diagram in log*(n) [this]	Maciej A. Czyzewski
[article]	Facet-JFA: Faster computation of discrete Voronoi diagrams [2014]	Talha Bin Masood, Hari Krishna Malladi, Vijay Natarajan
[article]	Jump Flooding in GPU with Applications to Voronoi Diagram and Distance Transform [2006]	Guodong Rong, Tiow-Seng Tan

Implemented Algorithms

	JFA*	JFA+	JFA
used improvement	noise+selection	noise	--
num. of needed steps	log*(n)	log4(p)	log2(p)
step size	p/(3^i)	p/(2^i)	p/(2^i)
research	(our)	(our)	[Guodong 2006]

Installation & Example

Project can be installed using pip:

$ pip install -e git+git://github.com/max-uho/fast_gpu_voronoi.git@master#egg=fast_gpu_voronoi

Here is a small example to whet your appetite:

from fast_gpu_voronoi       import Instance
from fast_gpu_voronoi.jfa   import JFA_star
from fast_gpu_voronoi.debug import save

I = Instance(alg=JFA_star, x=50, y=50, \
        pts=[[ 7,14], [33,34], [27,10],
             [35,10], [23,42], [34,39]])
I.run()

print(I.M.shape)                 # (50, 50, 1)
save(I.M, I.x, I.y, force=True)  # __1_debug.png

Development

If you want to contribute, first clone git repository and then run tests:

$ git clone [email protected]:maciejczyzewski/fast_gpu_voronoi.git
$ pip3 install -r requirements.txt
$ pytest

Results

Our method	Current best
JFA*	JFA
steps = log*(2000) = 4	steps = log(720) ~= 10

...for x = 720; y = 720; seeds = 2000 (read as n = 2000; p = 720).

Thanks