
|
Inverse
Dynamics [French]
Cédric Syllebranque and Samuel
Boivin
In AFIG 05
Abstract: The main goal of these works is to
compute the parameters of a dynamic model directly from images or from a
specific device (like the Kawabata machine).
In this
paper, we show how to extend an existing technique to compute the parameters
of a mass-spring system directly from synthetic images using a well-know 3D
modeling software. We iteratively compute the traction, bending and shearing
parameters by comparing the videos generated by this software and ours. We
use the final estimated parameters to produce a realistic synthetic image
sequences in Computer Graphics.
We are
also currently working on a new technique that estimates the Young modulus
and the Poisson coefficient directly from a real video of a solid getting
deformed under external action of a user. The external forces are captured
through a specific device that we built for this purpose. The parameters are
iteratively estimated using a new error metric for comparing a set of images.
Cloth
Simulation
Cyril Ngo-Ngoc and Samuel
Boivin
INRIA Research Report #5099
Abstract: This research report describes a
cloth simulation system including nonlinear behaviors. Our cloth model
simulates the fabric as a complex yarn interlacing structure. This approach
allows us to deal with the yarn scale problem by including the nonlinear
interactions happening inside this structure. This model has three different
components: the traction, the bending and the shearing. We also add new
friction terms in order to reproduce the yarn interlacing structure and the
nonlinear properties of a real cloth.
Moreover, this representation avoids a detailed 3D geometric model of
yarns which is almost always unusable in computer graphics.
Furthermore, our physical model is based on the Kawabata Evaluation System (KES). It reproduces the behavior of a
cloth using the parameters coming from direct measurements on real clothes.
Therefore we propose an identification procedure to compute all the physical
parameters of our model from the KES curves. Since many textile manufacturers
use this KES to design their fabrics, we are able to accurately simulate
their real fabrics in computer graphics using the manufacturer's parameters.
The simulation of a complete garment made of these fabrics could also be
achieved in a virtual fashion show to visualize its characteristics and/or
flaws for example.
|

|
Samuel Boivin
To be submitted
somewhere, one day...
Abstract: In this
work, we present our post-doctoral results on inverse rendering, to recover
all the reflectance parameters of a refractive medium from a single image. We
describe many different approaches for estimating the parameters of a typical
refractive object (diffuse transmittance, specular transmittance, refractive index, ...). We demonstrate that some intuitive ideas
produce wrong results and we show many different reconstructed scenes
containing a glass of wine for example. Our technique is able to estimate the
properties of these complex media using one or several images. Until now,
this technique has only been tried on various computer graphics scenes as
datum images.
|
|
Samuel Boivin
In ACM SIGGRAPH
2002 Course Notes #39
Acquiring Material
Models Using Inverse Rendering
Steve Marschner and
Ravi Ramamoorthi
Abstract: In this
course, we present our works on inverse rendering, to recover all the
reflectance parameters from a single image. We show some new extensions
regarding experimental validation and inverse rendering of refractive media
generating caustics.
|
|
Samuel Boivin and André Gagalowicz
In CGIV'2002
Proceedings
Abstract: In this paper, we
present an extension of our works, introducing resent a new a new
experimental validation of our results . Our method
is used to recover an approximation of the bidirectional reflectance
distribution function (BRDF) of the surfaces present in a real or synthetic
scene. This is done from a single photograph and a 3D geometric model of the
scene. The result is a full model of the reflectance properties of all
surfaces, which can be rendered under novel illumination conditions with, for
example, viewpoint modification and the addition of new synthetic objects.Our
technique produces a reflectance model using a small number of parameters.
These parameters nevertheless approximate the BRDF and allow the recovery of
the photometric properties of diffuse, specular, isotropic or anisotropic
textured objects. The input data are a geometric model of the scene including
the light source positions and the camera properties, and a single captured
image. We present several synthetic images that are compared to the original ones.
Some of the recovered reflectance values are compared to the real ones used
to produce an original synthetic image. Possible applications are given in
augmented reality such as novel lighting conditions and addition of synthetic
objects.
|
|
Samuel
Boivin and
André Gagalowicz
In ACM
SIGGRAPH 2001 Proceedings
[Note:
symbols in figure 3 page 110 are misprinted: the "rho" symbol is
missing before subscripts "d" and "s"]
Abstract: In this
paper, we present a new method to recover an approximation of the
bidirectional reflectance
distribution
function (BRDF) of the surfaces present in a real scene. This is done from a
single photograph and a 3D geometric model of the scene. The result is a full
model of the reflectance properties of all surfaces, which can be rendered
under novel illumination conditions with, for example, viewpoint modification
and the addition of new synthetic objects. Our technique produces a
reflectance model using a small number of parameters. These parameters
nevertheless approximate the BRDF and allow to recover
the photometric properties of diffuse, specular, isotropic or anisotropic
textured objects. The input data are a geometric model of the scene including
the light sources positions and the camera properties, and a single image
captured using this camera.
Our
algorithm generates a new synthetic image using classic rendering techniques,
and a lambertian hypothesis about the reflectance model of the surfaces.
Then, it iteratively compares the original image to the new one, and chooses
a more complex reflectance model if the difference between the two images is
greater than a user-defined threshold. We present several synthetic images
that are compared to the original ones, and some
possible applications in augmented reality.
|
|
Samuel Boivin
Thèse de l'École Polytechnique, 2001
Résumé: L'image de synthèse a souvent pour objectif de créer une illusion de la
réalité, et elle est aidée en cela par des algorithmes de rendu réaliste.
Malheureusement, peu de techniques se réfèrent à des images réelles pour
calculer des images synthétiques, limitant ainsi le réalisme des résultats
obtenus.
Nous proposons ici une nouvelle approche issue du rendu
inverse, et qui consiste à créer une nouvelle image de synthèse photoréaliste,
depuis une image réelle prise avec une caméra quelconque, et en utilisant un
modèle géométrique 3D de la scène (incluant les sources de lumière). Notre
méthode repose sur l'emploi d'une image unique, et procède à l'estimation
totalement automatique depuis cette image des paramètres d'une fonction de
distribution de réflexion bidirectionnelle (BRDF), fondée sur le modèle de
Ward. Nous démontrons que nous pouvons ainsi retrouver les réflectances des
surfaces diffuses, spéculaires (parfaites ou non), ainsi qu'isotropes,
anisotropes ou texturées, grâce à tout un ensemble d'algorithmes itératifs et
hiérarchiques. Chacun de ces algorithmes recherche la meilleure approximation
possible de la fonction de réflexion de la surface analysée, en minimisant
l'erreur entre l'image réelle et l'image synthétique générée par rendu
réaliste. De nombreux exemples de reconstruction géométrique et photométrique
sur des scènes d'intérieur sont montrés, ainsi que les images d'erreurs entre
l'image réelle et l'image synthétique.
Par ailleurs, nous apportons dans cette thèse plusieurs
idées théoriques et pratiques pour la conception d'un logiciel de calcul
d'images de synthèse qui soit à la fois rapide, mais aussi capable de créer
des images photoréalistes, alors que ces deux objectifs sont antinomiques. Au
delà de ces techniques de rendu réaliste et de rendu inverse, nous proposons
enfin plusieurs applications directes de notre méthode, pour la réalité
augmentée par exemple, ou la compression de séquences d'images.
|

|
Samuel Boivin and André Gagalowicz
Chapter of the book Imaging
and Vision Systems: Assessment and Applications, 2001
Jacques Blanc-Talon and Dan Popescu
Abstract:In this chapter, we
propose a new approach to compute the 3D geometry and the BRDF of surfaces
from real photographs. We use a pair of stereoscopic images (left and right)
to estimate both geometrical and photometric properties of objects inside a
real scene. The vision process recovers the 3D geometrical model, and the
properties of the camera including its position. The photometric process
estimates the reflectances of objects, limited to the perfectly diffuse case
and to the perfectly specular case. Several reconstructions are shown from
real photographs, including a complex interior scene. Some augmented
applications have been realized such as the generation of new synthetic
images from a novel viewpoint for example.
|
Coming soon...
|
Samuel Boivin and André Gagalowicz
In ERCIM News No.44, 2001
Abstract: Several techniques
have been recently developed for the geometric et photometric reconstruction
of indoor scenes. We have designed a new analysis/synthesis method that can
recover the reflectances of all surfaces in a scene, including anisotropic
ones, using only one single image taken with a standard camera and a
geometric model of the scene. Several industrial applications are now
possible using this advanced technique.
|

|
Samuel
Boivin, André
Gagalowicz and Youichi Horry
In International Journal
Machine Graphics and Vision, 2000
Abstract: This paper describes a new set of
techniques based upon laws of object composition for the modeling and
animation of deformable objects. They present some similarities with logical ,or morphological operations. This method is
unique because the volume of the merged object is held constant with respect
to its constituents. Composition from two objects is allowed for general
shapes which may have very different topologies. These techniques are applied
towards shapes which are approximated by a polyhedral description. When
integrated into a modeler, they can be used as modeling tools, animation
tools, deformation tools or even morphing tools depending on how they are
handled. One of the advantages of these tools is that the user may easily
deduce the procedure to follow in order to obtain a given result.
|
|
Rendu Réaliste par Radiosité [French]
Rapport de D.E.A. de l'Université de Technologie de Compiègne(U.T.C.), 1995
Résumé: Ce rapport de DEA
décrit tout d’abord un état de l’art conséquent sur les méthodes
existant en radiosité et en rendu photo-réaliste. Il contient ensuite le
descriptif très détaillé d’un logiciel de rendu réaliste innovant et
développé à l’INRIA Rocquencourt.
|
Some plants coming soon...
|
Simulation
de Croissance de Végétaux [French]
Samuel Boivin
Rapport de Maitrise de l'Université de Reims/Université
Paris VIII, 1994
Résumé:Ce rapport décrit
mon stage de Maîtrise effectué sur la simulation de croissance de végétaux.
J’ai implémenté mon propre moteur de rendu basé sur un Z-Buffer (les
cartes graphiques grand public de l’époque ne disposant pas encore de
cet algorithme en hard). J’ai également développé mon propre simulateur
de plantes et d’arbres en synthèse d’images en utilisant des
L-systems, et en me servant de mon logiciel de rendu pour calculer les
images.
|
|
A Rendering Method
For the Realistic Simulation of Natural Scenes
Samuel Boivin and Laroussi Doghman
Abstract: In this paper we describe a new rendering
algorithm to produce photo-realistic images. This software was the precursor
of the much more evolved rendering software called Phoenix that I have
developed during my phD. Phoenix is used to render
all the CG images since 2001.
|