Call for participation (DIMACS Implementation Challenge on Surface Reconstruction)

tamal dey tamaldey at
Fri Oct 25 16:01:28 PDT 2002

DIMACS Implementation Challenge on Surface Reconstruction

 Organizers: Claudio T. Silva (csilva at
 	     Tamal K. Dey (tamaldey at

Recent technological and algorithmic advances have improved the
process of acquiring discrete sample points from a 3D geometric
object. This data acquisition includes scanning the boundary
of a real world 3D object with a variety of scanners as well as
discrete data points generated by some scientific simulation.
Several algorithms have been proposed in recent years to
approximate the surface from which the samples are acquired.
The ultimate proof of the success of these surface reconstruction
algorithms depend on their performance on real data that pose
difficulty due to noise, undersampling and size. 


The goal of the Implementation Challenge is to determine 
how algorithms depend on the
structure of the underlying data sets, and to determine realistic
algorithm performance.


We invite research groups to participate in this Implementation Challenge. 
Participants can provide their own data sets which will be added to a 
pool of test data sets including our own data sets and those provided
by all participants.  A sample of those
datasets will be available from our web site (listed below). 
Interested person/groups should send their intention in participating
in the Challenge by sending an email to the organizers.


The results of this challenge will be reported in a session of the
DIMACS workshop on surface reconstruction, April 30-May 02, 2003 organized
by Nina Amenta and Fausto Bernardini.
We will report different results, including 
(i) Machine performance:  speed, memory usage
(ii) Output quality: output size, nonmanifold-ness, holes
(iii) Scope: Generality. 
Ideally the codes should run on a PC running Linux and/or Windows 
with 1GB of RAM, and 120 GB of disk space (single disk).


The input datasets will consists of a combination of formats,
including unorganized points (without normals), points with normals,
and collections of (registered) range scans, and they will range in
size from small (few hundred samples) to large (few million samples).
(Algorithms do not necessarily need to work on all the data.)


The output of the softwares should be in Geomview OFF format.
We will annouce other acceptable formats as we hear from  
the participants. 

Web site (tentative):

The compgeom mailing lists: see
or send mail to compgeom-request at with the line:
send readme
Now archived at

More information about the Compgeom-announce mailing list