Abstract
In recent multi-view photometric stereo systems that captures live performance, reconstructed models are hole present and temporally incoherent due to occlusion, inaccurate correspondence and warping error. Using Inter-frame information is crucial to complete model geometry and ensure temporal coherence. We attempt to address these problems by applying a number of techniques within a temporal window, including adaptive local ICP, CSRBF deformation, thin plate spline and Poisson surface reconstruction. Analysis about these techniques is provided and some promising results in one of the challenging cases are shown. Also, we demonstrate how color and normal images can be incorporated to generate globally consistent texture with enhanced detail on the reconstructed model.
Abstract
The Laplace operator is a fundamental geometric object and has many
desirable properties. In addition, the Laplace operator is intimately
related to heat diffusion process on a manifold, relating geometry of
a manifold to the properties of the heat flow. In this talk, I will
present our work on deriving multiscale shape signatures based on
heat diffusion, and on clustering of biomolecular conformations based
on the eigenvectors of Laplace operator. I will also talk about some
future directions on both shape and data analysis that interest me.
Abstract
Symmetry is one of fundamental characteristics of many natural and man-made objects. Several algorithms have been proposed recently to extract symmetry descriptors for partial and perfect symmetries. These data were used for shape symmetrization, hole-filling, viewpoint selection, and more. In this work we are trying to develop a general framework which would allow user to define target shape or image directly in symmetry space. This allows transferring symmetric properties from one object to another, symmetrizing images and filtering/editing images in symmetry space.
Abstract
We present a multiple-feature matching approach for automatically matching small fragments of archaeological artifacts such as Bronze-Age and Roman frescoes. Our approach extends the traditional collection of 2D image and 3D geometry based features by incorporating high resolution 2D normal maps. In addition, we introduce a set of domain specific feature descriptors motivated by the visual cues used by archaeologists for reassembly. These features may be computed at low cost and correlate directly to the objects’ physical characteristics. In contrast to current approaches, we utilize machine learning techniques to evaluate the effectiveness of our descriptors, and to train our system to identify new matches. We have tested our system on three datasets of scanned fresco fragments, experimenting with both all-feature matching and automatic feature selection.
Abstract
The work on design exploration is based on three exemplary design explorations conducted by Axel Kilian during the
course of a PhD at the Massachusetts Institute of Technology. The three examples represent three different types of
explorations. One, a chair design, shows exploration as a fine tuning exercise of bringing together multiple design
constraints to achieve a well balanced design overall. Two, illustrated with a concept car design study, shows
design exploration with the focus on innovation, through identifying and assembling design constraints step by step
to define the design challenge. And three, using a form finding tool, shows design exploration as a process of
discovering new design through a purpose built software piece that captures all known constraints. Of course there
are many more types of explorations not captured by these categories but they help to shed light on what remains
one of the biggest challenges in computational design, supporting design beyond recording of the designers intend
to a stage where both formal and conceptual variations of design ideas can be fluently explored at similar speeds
as in a brainstorming session between people, yet make use of computation to extend the depth at which designs are
being considered, and allow for the discovery of new ones possibly missed without computational support. Bio
Axel Kilian studied Architecture at the University of the Arts Berlin and specialized in Design and Computation for
a Masters of Science, a Ph.D., and a post-doctorate at the Massachusetts Institute of Technology. He has been
teaching workshops in computational design and design studios and has lectured widely on the topic of generative
design at universities, in architectural practices and as a tutor in the smart geometry group workshop series since
2003. >From 2007 to 2009 he was an Assistant Professor in Design Informatics at Delft University of Technology.
Since 2009 he is an Assistant Professor for Computational Design School of Architecture of Princeton University.
|
