Eurographics Italian Chapter Conference (2008)

V. Scarano, R. De Chiara, and U. Erra (Editors)

MeshLab: an Open-Source Mesh Processing Tool

P. Cignoni, M. Callieri, M. Corsini, M. Dellepiane, F. Ganovelli, G. Ranzuglia

Visual Computing Lab, ISTI - CNR, Pisa Italy

Abstract
The paper presents MeshLab, an open source, extensible, mesh processing system that has been developed at
the Visual Computing Lab of the ISTI-CNR with the helps of tens of students. We will describe the MeshLab
architecture, its main features and design objectives discussing what strategies have been used to support its
development. Various examples of the practical uses of MeshLab in research and professional frameworks are
reported to show the various capabilities of the presented system.
Categories and Subject Descriptors (according to ACM CCS): I.3.3 [Computer Graphics]: Line and Curve Genera-
tion

1. Introduction Another more serious issue of these system is their very

short life expectation: maintenance is neither a requirement
3D modeling and editing tools are a quite common nowa-
nor a rewarding activity for a researcher, so in most of the
days, and the software market offers a very vast set of solu-
cases these tools, after a few months, often lie in a rather
tions for many different purpose, targeting the most variable
abandoned state.
audience, so the idea of developing yet another tool should
be evaluated with a bit of care. On the other hand there is Another source of research-level F/OSS systems are gov-
still space for specific, sharp tools that offers functionalities ernment founded projects where, in some cases, the funding
rarely available (particularly when they are free). In this pa- organization encourages the dissemination of the results un-
per we present MeshLab the 3D mesh processing system that der a liberal licensing scheme. In these cases stability and
we have developed with the help of many students in the last maintenance are guaranteed for the project duration, but at
two years and that fits in the peculiar niche of advanced re- the end of the project the developed software systems face
search tools useful for a wider audience. again the risk of being abandoned.
For our system we have decided try to reduce our com-
1.1. Motivations mitment in the project as much as possible and to leverage
on students workforce for the more practical implementing
There are many basic motivations for a research institution duties. This approach as many positive effects: reducing our
to start a free or open source software(F/OSS) project. Com- commitment left us more time for classical research activ-
monly the most frequent reason is to allow the so called re- ity (usually more rewarding from a professional point of
producibility of research results; in this case a research group view), binding the development of the project with a uni-
releases, under a more or less liberal license, a frozen version versity course has the consequence of guaranteeing willing
of the software system behind a given published work; while work force (the students has to complete a piece of Mesh-
this kind of approach should be strongly encouraged by jour- Lab as part of their course jobs) and it forces us to keep the
nal and conferences because allows to the reviewer to get a project updated from year to year.
better assessment of quality of the proposed solutions, from
a software engineering point of view the release software is
often far from being a clean and robust software product.
2. MeshLab general architecture
This is not a negative point of view from a scientific point
of view, because the purpose of the released software is well MeshLab was designed as a general 3D mesh processing
different from the one of being a real product. system tool with the following primary objectives in mind:

c The Eurographics Association 2008.
130 Cignoni et al. / MeshLab

• ease of use. The tool should be designed so that users just MeshLab and LaTeX, pdf files with 3D objects
without high 3D modeling skills could use it (at least for that can be seen with the current free version of Acro-
the most basic functionalities) bat Reader.
• deep of use. The tool should be designed so that advanced • Mesh Cleaning Filters:
users can tweak and extend it by adding functionality and – removal of duplicated, unreferenced vertices, null
or by modifying all the involved parameters. faces, small isolated components
• single mesh processing oriented. The system should try – coherent normal unification and flipping
to stay focused on mesh processing instead of mesh edit- – erasing of non manifold faces
ing and mesh design where a number of fierce competi- – automatic filling of holes
tors already crowd the software arena (notably blender, • Remeshing filters:
3D Max, Maya, and many others). – High quality edge collapse simplification (even with
• Efficiency. 3D scanning meshes can easily be composed texture coords preservation)
by millions of primitives, so the tool should be able to – Ball Pivoting surface reconstruction
manage them. – Subdivision surfaces (loop and butterfly)
• Sustainability. The project should be able to grow and – Feature preserving smoothing and fairing filters
self sustain itself for at least some years. • Various Colorization/Inspection filters
– Gaussian and mean curvature
As a result MeshLab presents itself a mesh viewer applica-
– Border edges, geodesic distance, from borders
tion, where a 3D object, stored in a variety of formats can be
– Non two-manifold edges and vertices
loaded and interactively inspected in a easy way, by simply
– Self intersecting faces
dragging and clicking on the mesh itself. MeshLab supports
– Ambient Occlusion. An ambient occlusion field can be
a ever growing variety of 3D formats (all the most commons
computed and stored per vertex
format are supported) to accommodate the broadest set of
users. Once loaded a mesh the user can work on it by mean • Measuring tool. You can take linear measures between
of a large set of direct parametric filters, that performs unat- points of the displayed meshes
tended automatic task like smoothing, re-meshing or sim- • Slicing tool. A new tool that allows to export planar sec-
plifying, or by mean of interactive tools. Figure 1 shows tions of a mesh in SVG format
an example of an interactive filter when the user drag the • 3D Scanning tools
– Alignment: ICP based range map registration tool, for
mouse over the mesh a local smoothing is performed on the
putting meshes into the same reference space.
touched portion of the mesh in real time; in this case the
– Merging of multiple meshes the Poisson surface recon-
result is that the user is washing out some features of the
struction source code
object. No classical design-oriented features are provided,
• OpenGL Shader based rendering
structured editing of complex scene graphs is not supported
• Large rendering. Images up to 16k x 16k pixels can be
by design. Multiple meshes can be loaded together and pro-
created for high quality poster printing
cessed separately or together following a flat approach based
on layers. This approach is, in some sense, a bit similar to the
classical raster processing tools that focus on the processing 2.1. MeshLab Development
of a bitmap and their "global" composition rather focusing One of the MeshLab ambitious goals was to create an open
on the pure painting operations. source application that is backed up by a sustainable devel-
The mesh processing functionalities that MeshLab cur- opment model. As previously discussed a common prob-
rently provide are many and a short, incomplete, high level lem of many open source tool is that either they are small
list of MeshLab characteristic is presented in the following: enough to be created and maintained by a small team, or,
if they require a considerable development effort, the devel-
• Interactive selection and deletion of portion of the mesh. oper team need to be economically supported in some way.
Even for large models. Many times the projects originates as research projects that
• Painting interface for selecting, smoothing and coloring are freed/opened (or simply released under some OSI ap-
meshes. proved license) at the end (or during) the (financially sup-
• Input/output in many formats: ported) research project. In these case the big incognito is the
– import: PLY, STL, OFF, OBJ, 3DS, COLLADA,
long term maintenance of the tool. In some cases the origi-
PTX, X3D, VRML
nal developers, mostly for ethical and sentimental reasons,
– export: PLY, STL, OFF, OBJ, 3DS, COLLADA, X3D,
continue a low priority maintenance the software correcting
VRML, DXF
small bugs and applying minor modifications, but great im-
– Point Clouds support. 3D files that are composed only
provements are difficult to carry on. A direct consequence of
by points are well supported in PLY and OBJ format.
this situation is that many many interesting projects lie in a
– U3D support; MeshLab is the first open source tool
rather abandoned state.
to provide direct conversion of 3D meshes into the
U3D format. with this format users can create, with While we could base the development of the system on a

c The Eurographics Association 2008.
 Cignoni et al. / MeshLab 131

should be included in the core of the system. And this can

cause the some unpredictable delays in the development of
some (possibly very important) features.
As a result MeshLab started out in the autumn of 2005 and
most of the 60k lines of code composing its core, have been
written by willing students as part of their course work; each
year students are adding new and new features to the system
and they are actually making it to grow at a stead rate. The
long list of many contributors to the MeshLab base code is
well shown in the home page of the system.

3. MeshLab and Arc3D web service

As an example of the capabilities of MeshLab we briefly
Figure 1: A typical snapshot of MeshLab in action: the mesh report how MeshLab covered a critical role in a publicly ac-
under processing is interactively displayed and the user can cessible framework for the 3d reconstruction from sequence
work on it by mean of a large set of unattended parametric of photos. Two partners of the Epoch Network of Excel-
filters, or by mean of interactive tools, like the one shown lence [Epo06], the ESAT-PSI lab of K.U.Leuven (Belgium)
in the above image, where the user is smoothing out some and the Visual Computing Lab of CNR-ISTI (Italy) have set
features of the object with some simple mouse strokes. up a low-cost 3D reconstruction pipeline to be used in the
cultural heritage field. The idea of the designed system is
that only a digital photocamera, a pc, and an Internet con-
nection are necessary for a user to reconstruct scenes in 3D.
large stable C++ library for mesh processing task [The04],
The central part of this approach is a web accessible service,
the whole task of building a consistent interface to present
called Arc3D that allows to reconstruct raw 3D data directly
the library functionalities to the user (and add new missing
from photo sequences, whose results are then processed by
pieces) was considerable.
MeshLab into clean usable 3D models.
Very large projects that can afford the support of a large
A more detailed description of the complex inner working
community can rely on the loose sustain of many individ-
of this system, and how it magically succeed to reconstruct
uals, but large communities are difficult to be constructed.
the basic 3D data from simple sequences of photos, can be
We decided for the above reasons for a very modular archi-
found in [VVG05]; in this section we mostly describe the
tecture and to subdivide the whole system into as small and
second part of the whole pipeline the one that brings to raw
as independent as possible pieces. These final pieces were
reconstructed data to 3D objects. For accessing to the ser-
small enough to be assigned, as course tasks, to students of
vice, downloading the uploading software add trying the re-
FGT a Computer Graphics course at the Computer Science
construction service visit [Arc07].
Faculty of the University of Pisa.
From a practitioner point of view, MeshLab is assembled
3.1. MeshLab processing of Arc3D data
by a central skeleton framework with almost no capability
and a large set of independent plugins that implement all the The output of the Arc3D web service is constituted by a set
functionalities of meshlab, from loading the many different of range maps with texture and a per-pixel quality measure
file formats, to simple filtering like cleaning and smoothing that denotes the number of found matches among all the
algorithms to complex interactive tools that directly act over images. For each range map all the intrinsic and extrinsic
the mesh under the control of the user like the painting tools camera parameter are available, that makes easy to build, for
or the alignment subsystem. each range map a simple point clouds with the points placed
in the right spatial position.
The idea proved successful, motivating the students much
more than classical ’mock-up’ projects usually faced during Like all the data coming from digital acquisition devices
courses, moreover the fact that their code would have been they are affected in some measure by various kind of noise
actually used in a real project used by real people (we usu- and errors. Knowing the kind of issues that are in the ac-
ally show to the students people around the world that dis- quired data can help to remove or at least reduce them.
cusses MeshLab bugs and workaround) stimulated them to
MeshLab provide a specific tool for importing the data
take their task with greater serious attention.
produced by the Arc3D service. The tool is integrated in-
On the other hand not all the students are equal, and for side meshlab as an input filter, when the author try to open
tasks that where assigned to low scoring students, the results a ".v3d" file created from the Arc3D the interface shown in
are pieces of code that not always was decided that they Figure 2 is displayed to the user. All the images used for the

c The Eurographics Association 2008.
132 Cignoni et al. / MeshLab

the error is more or less uniform along all the directions.

Moreover the subsampling process is done

3.3. Outlier management

From a practical point of view it is convenient to consider
also procedures for the management of outliers. For a vari-
ety of reasons (wrong matching, non correct lighting setup,
difficult surface reflectance properties) some of the acquired
points can represent ghost surfaces that do not belong at all
to the original surface. Luckily enough these points are usu-
ally rather isolated, or in other words they do forms small
floating pieces far form the main surface. MeshLab provides
a couple of automatic filters that allows to cope with these
issues, these filters removes the all the small isolated con-
Figure 2: The MeshLab interface of the importer for photo- nected components that are smaller than a chosen size ex-
reconstrcuted data from the Arc3D service. A synthetic color pressed either spatially or in terms of primitives compos-
map depicting the quality of each range map is shown for ing the component. Figure 3 shows an example of the fil-
each reconstructed photo. A number of option allow to fine ters that automatically removes the small isolated pieces that
tune the processing of the chosen range maps. float around each single range map. The same range maps is
shown before (left) and after (right) the processing.

3.4. Noise management

reconstruction process are shown on the right with a color
coded measure of the quality of the reconstruction process. Image based acquisition methods suffers of noise problems
The user can select some of the images (usually the best ones that are in some way the opposite of the one that happens
and /or the ones that cover in the better way the targeted ob- in traditional active scanning technologies. In fact the most
ject) and the 3D data generated from the Arc3D is used to difficult surfaces for passive techniques are the ones that are
build a set of range maps ready to be utterly cleaned inside very flat, uniform and smooth. where the absence of promi-
MeshLab and then integrated in a single clean object. The nent visible distinctive features, make difficult any match-
user can also paint individual masks over the image for dis- ing procedure. The direct consequence of this behavior is
carding portions of the images. A number of parameters are that flat smooth surfaces can be affected by a considerable
present in the left side to carefully drive the process of con- amount of noise. On the other hand it is possible to partially
verting the raw 3D data into clean high quality range maps. cure this problems by analyzing the original images used for
Once the range maps have been imported a set of different the reconstruction process. In fact by detecting the portions
operations have to be performed on them (cleaning, merg- of the image that do not posses enough features and marking
ing, aligning) in order to obtain a nice, ready to be used 3D them it is possible to apply them in a latter stage a heavier
model. In the next paragraphs we discuss some aspects of smoothing pass.
these processing steps
Moreover sometimes, typically when the photos are not
taken in the perfect setting, the boundaries of high jumps
3.2. Weighted averaging and depth aware triangulation region can present larger errors with eventually large and
coherent outliers. Also these situations can be detected and
First of all, some subsampling steps are usually necessary, corrected almost automatically by searching the boundary
current digital cameras has very resolute CCD an can eas- regions and applying a erosion-dialation approach that re-
ily generate tens of megapixel images. While this kind of moves one or more strip of pixels from the dangerous jump
definition is useful from the reconstruction point of view be- zones.
cause it helps the robustness of the feature matching parts (at
Once you have removed all the outliers you have to choose
high definition even flat surfaces have some kind of texture),
if you should fill all the created holes and interpolate in some
when converting these maps to 3D meshes this resolution is
way the original mesh, This can be done inside meshlab, that
often not very useful. Infact the spatial accuracy of the re-
provides some hole-filling filters that can close all the small
constructed data is not uniform along the axis: along the xy
holes under a given size. Thy applies a variant of the algo-
image plane there is much more precision than along z depth
rithm presented in [Lie03], with various different heuristics
direction. When you will integrate the various meshes the fi-
for choosing the
nal quality will be affected by an error dependent from the
maximal erorr of the single range-[maps/meshes so it is a Figure 4 shows the combined application of the tech-
good suggestion to subsample the images at the point that niques here described, large outliers and dangerous borders

c The Eurographics Association 2008.
 Cignoni et al. / MeshLab 133

Figure 3: MeshLab provides filters that automatically removes the small isolated pieces that float around each single range
map. The same range maps is shown before (left) and after (right) the processing.

Figure 4: The figure shows the combined application of erosion dilation approach for cleaning out the depth jumps outliers
that sometimes appears. A set of small holes, that remains from outliers removal, is covered automatically by robust holefilling
technique.

(look above and around the bull statue) are removed. A set step but range map from each sequence get out already well
of small holes (all over the wall), that remains from previous aligned from the Arc3D service. The aligning issue arise
outliers removal, is covered automatically too. when you wont to join two or more different sequences. In
this case you have to align them together.
4. MeshLab as a general range map processing tool The Alignment process adopts a classical approach based
on first, a pairwise local and then a global alignment [LR01,
One of the interesting characteristic of MeshLab is the
Pul99]. The initial placement placement of each mesh is
presence of all the basic pieces of the 3D scanning soft-
driven by the user and require the selection of four or more
ware pipeline. Infact beside the rangemap cleaning tools de-
points on both the meshes. This initial step is followed by
scribed in the previous sections, that were tailored for the
an automatic process that fine tunes the whole alignment
noisy meshes output from multistereo reconstruction tech-
. The alignment code used in meshlab is a derivation of
niques, but works quite well for any kind of range maps,
the one used in Scanning Tools of the Visual Computing
inside meshlab there are tools for the subsequent required
Lab [CCG∗ 03], that has been used in a number of projects.
alignment and merging processing steps.

4.1. Aligning 4.2. Merging

The aligning process it the step in which you take a set of The individual range maps, after having been carefully
different range maps of a same object, each one in its onw cleaned and prepared, are ready to be integrated and fused
reference space and you rototranslate them in a single con- in a single coherent model. All the range maps coming from
sistently aligned space. Usually in the traditional scanning a same sequence are already well aligned, so the next step is
pipeline [BR00], the aligning step, come before the merging, the one of to apply a surface reconstruction filter that fuses

c The Eurographics Association 2008.
134 Cignoni et al. / MeshLab

Figure 5: Three samples of the meshes that can be obtained by using the Arc3D web service reconstruction and the MeshLab
system. On the Left, a textured mesh representing one of the two lions in the Loggia della Signoria in Florence. On the center
an untextured statue from the Portalada in Ripoll, a large portal in Ripoll, near Barcelona, on the right a reconstruction of the
Paris arc du triomphe, untextured for better evaluation of the geometric quality of the geometric shape. The ambient occlusion
lighting used to better enhance the shape features was computed with MeshLab.

Figure 6: A snapshot of the MeshLab Aligning tool, that allow to register different range maps, or in the case of data coming
from the Arc3D web service, portions of a same object that have been reconstructed from different photo sequences.

all the meshes in a single new mesh integrating the various more safe approach. Figure 5 shows three samples of the
parts together in a seamless way. MeshLab offers three dif- reconstruction process. On the Left, a textured mesh rep-
ferent surface reconstruction algorithms. The first one is a resenting one of the two lions in the Loggia della Signoria
interpolatory triangulation filter based on the Ball-Pivoting in Florence, the color was integrated from the many photos
algorithm [BMR∗ 99], that tries to build a surface connect- directly during the reconstruction. On the center an untex-
ing all the input points. This kind of algorithms do not work tured statue from the Portalada in Ripoll, a large portal in
very well in presence of very noisy input data like the one Ripoll, near Barcelona reconstructed using the Poisson sur-
coming from the Arc3D service. The other two are implicit face reconstruction algorithm; a watertight surface was ro-
surface approaches that uses the input data to build a im- bustly constructed, even if the input data contained only in-
plicit representation and then polygonalize it using marching formation on the front of the statue, building an interpolating
cubes [LC87] variants. One is based on the Poisson surface surface even for the back of the model. On the right a recon-
reconstruction algorithm [KBH06], and it uses part of the struction of the Paris arc du triomphe, done again with the
original code provided by the authors themselves, and the plymc apprach untextured for better evaluation of the geo-
second one, called plymc, developed at the Visual Comput- metric quality of the geometric shape. The ambient occlu-
ing Lab and described in [CCG∗ 03], is an extension of the sion shading that is used to better enhance the shape features
approach of [CL96]. The Poisson based surface reconstruc- was computed with MeshLab.
tion algorithm has the very nice characteristic that always
build a watertight hole-free surface filling with an interpola-
tory surface all the missing parts, but currently do not sup- 5. MeshLab as an inspection, verification, and
port the color preservation. On the other hand the plymc ap- assessment system
proach preserves the color during the processing but leaves Last but not least we would remember the many visualiza-
the unsampled areas as holes in the mesh; from a Cultural tion, inspection and healing features of MeshLab. One of the
Heritage point of view this behavior could be considered a common problem in managing real world 3d meshes is that

c The Eurographics Association 2008.
 Cignoni et al. / MeshLab 135

common models are often plagued by many different issues,

and inconsistencies both topological and geometrical. Typi-
cal example of topological issues are the presence of vertices
that are not referenced, or duplicated among faces, incon-
sistency in the face orientation, edge and vertex non mani-
foldness. For all of them meshlab provide tools for detect-
ing and in many cases removing them. Similarly there are
many geometrical issues that can arise frequently, like the
plain presence of self-intersecting faces, ghost geometry in-
side the object, small floating pieces around the mesh, badly
shaped triangles, fold-over and other high curvature config-
urations. Again MeshLab provide tools for detecting these
situations. In figure 8 two example of the visualization tools
are shown. On the left the mesh is transparently rendered
Figure 7: Smoothing filters can help the process of noise re-
with a color depending on the surface curvature; rendering
moval. Small bumps due to noise from the acquisition device
the mesh with an x-ray effect allow to easily detect the pres-
can be automatically reduced.
ence of extraneous geometry inside the object itself. On the
right, self intersecting face are automatically detected and
colored in red.
An interesting unique feature of MeshLab is that it can on Visualization and Computer Graphics 5, 4 (Oct.-Dec.
export meshes in the U3D format. Meshes in this format 1999), 349–359.
can easily included in PDF files using pdfLaTeX and the [BR00] B ERNARDINI F., RUSHMEIER H. E.: 3D Model
movie15 latex package. Acrobat Reader supports the inter- Acquisition. In Eurographics 2000, State of the Art Re-
active viewing of these objects since version 7; click on the ports Proceedings (Aug. 24–25 2000), Eurographics As-
figure 9 to interactively rotate and zoom the Gargoyle. It is sociation, pp. 41–62.
worth noting that the u3d format is quite compact: the u3d [CCG∗ 03] C ALLIERI M., C IGNONI P., G ANOVELLI F.,
model of the displayed gargoyle is 130k, less than the png M ONTANI C., P INGI P., S COPIGNO R.: VCLab’s tools
file used as a placeholder before activating the 3D model or for 3D range data processing. In VAST 2003 (Bighton,
for very old pdf browsers that do not support 3D objects. UK, Nov. 5-7 2003), D. Arnold A. C., Niccolucci F.,
(Eds.), Eurographics, pp. 13–22.
6. Conclusions and Future Works [CL96] C URLESS B., L EVOY M.: A volumetric method
The system has proved a success over any initial prediction. for building complex models from range images. In
The last version has been downloaded 10000 times in the Comp. Graph. Proc., Annual Conf. Series (SIGGRAPH
first three weeks, there are thousands of users from all the 96) (1996), ACM Press, pp. 303–312.
world with at least 600 users that have used it for opening [Epo06] E POCH: The European Network of Excel-
more than one hundred of meshes, Users come from hun- lence on ICT Applications to Cultural Heritage (IST-
dreds of universities and renowned commercial firms that 2002-507382). More info on: http://www.epoch-
have found MeshLab useful in contexts different from the net.org/, 2006.
original one of Cultural Heritages.
[KBH06] K AZHDAN M., B OLITHO M., H OPPE H.: Pois-
Acknowledgments We acknowledge the financial sup- son surface reconstruction. Proceedings of the fourth Eu-
port of “EPOCH" (EU Network of Excellence, IST-2002- rographics symposium on Geometry processing (2006),
507382). A warm thank you to the many developers and 61–70.
code contributors that have helped to build the current Mesh-
[LC87] L ORENSEN W. E., C LINE H. E.: Marching cubes:
Lab system, to the Arc3D team, and, particularly, to M.
A high resolution 3D surface construction algorithm. In
Kazhdan and M. Bolitho for the Poisson surface reconstruc-
ACM Computer Graphics (SIGGRAPH 87 Proceedings)
tion code.
(1987), vol. 21, pp. 163–170.
[Lie03] L IEPA P.: Filling holes in meshes. In EG/ACM
References
Symposium on Geometry Processing (SGP) (June 2003),
[Arc07] Arc3d: Automatic reconstruction conduit. More Eurographics, pp. 200–206.
info on: http://www.arc3d.be/, 2007. [LR01] L EVOY M., RUSINKIEWICZ S.: Efficient variants
[BMR∗ 99] B ERNARDINI F., M ITTLEMAN J., RUSH - of the ICP algorithm. In Third Int. Conf. on 3D Digital
MEIER H., S ILVA C., TAUBIN G.: The ball-pivoting al- Imaging and Modeling (3DIM 2001) (May 28th - June 1st
gorithm for surface reconstruction. IEEE Transactions 2001), IEEE Comp. Soc., pp. 145–152.

c The Eurographics Association 2008.
136 Cignoni et al. / MeshLab

Figure 8: Some of the many inspection tools inside meshlab. On the left the mesh is transparently rendered with a color
depending on the surface curvature; on the right, self intersecting face are automatically detected.

(gargoyle.u3d)

Figure 9: MeshLab can export meshes in the U3D format. Meshes in this format can easily included in PDF files using
pdfLaTeX. Acrobat Reader supports the interactive viewing of these objects; click on the above picture to interactively rotate
and zoom the Gargoyle.

[Pul99] P ULLI K.: Multiview registration for large 2004.

datasets. In Proc 2nd Int.l Conf. on 3D Digital Imaging [VVG05] V ERGAUWEN M., VAN G OOL L.: Web-based
and Modeling (1999), IEEE, pp. 160–168. 3D Reconstruction Service. Machine Vision and Applica-
tions 17, 6 (2005), 411–426.
[The04] T HE V ISUAL C OMPUTING L AB - ISTI - CNR:
Visual computing library: a gpl c++ library for mesh pro-
cessing. More info on: http://vcg.isti.cnr.it/,

c The Eurographics Association 2008.