FLUX(1) General Commands Manual FLUX(1)
NAME
flux - converge towards a target image using a Monte Carlo algorithm.
SYNOPSIS
flux [IMAGE]
flux [IMAGE] [options]
DESCRIPTION
Flux is a command-line tool that uses a Monte Carlo (MC) algorithm to
converge towards a target image using a random sequence of nearest
neighbor pixel swaps. Images produced by flux resemble iron filings
manipulated by a magnetic field.
Flux works by repeatedly attempting to swap the RGBA (red, green, blue,
alpha) values of two adjacent pixels. The algorithm computes the dif‐
ference in the magnitude of the two RGBA pixel vectors compared to the
same vector in the target image before and after the trial move and
accepts the move according to a Boltzmann factor. The system is coupled
to a thermal bath which allows the user to control the probability of
accepting moves that take the image away from the target, thus allowing
the rate of convergence to be tuned.
Flux avoids unneccessary dependency on zlib and libpng by using LodePNG
to decode and encode PNG images. MersenneTwister is used to provide
high quality and efficient random numbers.
Flux must be passed at least one command-line argument, the path of a
PNG image. Flux creates an initial canvas by randomizing the pixels of
the target image. As the algorithm progresses flux will output a
sequence of images showing the evolution of the the canvas. In addi‐
tion, the current acceptance rate is printed to stdout allowing the
user to monitor the rate of convergence.
Flux is trivially parellelizable, MPI support will follow (allowing
efficient processing of very large images).
OPTIONS
flux supports the following short- and long-form command-line options.
Invoking flux with no options will print a help message.
-h, --help
Print the help message.
-i INT, --iterations INT
Where INT is an integer specifying the number of iterations.
Iterations are measured in units of Monte Carlo sweeps, where
one sweep equals, on average, one trial move per pixel, i.e.
there are width x height iterations per sweep.
-f INT, --frames INT
Where INT is an integer specifying the number of frames to be
saved.
-t DOUBLE, --temperature DOUBLE
Specify the temperature of the thermal bath. A higher tempera‐
ture will mean that more trial moves are accepted. At zero tem‐
perature, only moves that take the canvas closer to the target
will be accepted (comparable to a hill climbing algorithm). Tem‐
perature adds noise to the system and can be used to tune the
rate of convergence as well as the style of image produced.
-l, --log
Use logarithmic spacing for sampling points. This is useful when
convergence is very slow.
-r, --reverse
Run flux in reverse mode. Start from the target image and evolve
to a disorded state.
-d DIRECTORY, --directory DIRECTORY
Specify a directory for the output files. DIRECTORY should be
in the working directory and will be created if it doesn't
already exist.
-m, --monitor
Turn on acceptance monitoring. If set, flux will monitor statis‐
tics for the acceptance rate and write data to log.txt
DEFAULTS
sweeps 10000
frames 1000
temperature 0.1
logarithmic off
reverse off
monitor off
SEE ALSO
rectify(1)
CREDITS
LodePNG, http://lodev.org/lodepng.
MersenneTwister, www.math.sci.hiroshima-u.ac.jp/~m-mat/MT/emt.html.
BUGS
Email bugs, comments, and artwork to lester.hedges+flux@gmail.com.
Lester Hedges 2013/01/22 FLUX(1)