KAUST Research Conference 2017:
Visual Computing – Modeling and Reconstruction

April 10 to April 12, 2017

Agenda

  • Day 1Monday, April 10th
  • Day 2Tuesday, April 11th
  • Day 3Wednesday, April 12th
8:00 am

Breakfast - Hall 1

KAUST 08:00 - 08:30 Details

8:30 am

Registration including meet and greet

KAUST 08:30 - 09:00 Details

9:00 am

Conference opening and Welcome remarks

KAUST 09:00 - 09:30 Details

WOLFGANG HEIDRICH, Director VCC
9:30 am

Keynote Lecture: The Power of Primal/Dual Meshes for Modeling and Animation

Triangle and tetrahedral meshes have found widespread acceptance in computer graphics as a simple, convenient, and versatile representation of surfaces and volumes. In particular, computing on such simplicial meshes is a workhorse in a variety of graphics applications. In this context, mesh duals (tied to Poincare duality and extending the well-known relationship between Delaunay triangulations and Voronoi diagrams) are often useful, from physical simulation of fluids to mesh parameterization. However, the precise embedding of a dual diagram with respect to its triangulation (i.e., the placement of dual vertices) has mostly remained a matter of taste or a numerical after-thought, and barycentric vs. circumcentric duals are often the only options chosen in practice. In this talk we discuss the notion of orthogonal dual diagrams, and show through a series of recent works that exploring the full space of orthogonal primal/dual meshes is not only powerful and numerically beneficial, but it also reveals (using tools from algebraic topology and computational geometry) discrete analogs to continuous properties. Applications varying from point sampling and fluid dynamics, to barycentric coordinates and self-supporting masonry will be covered.

KAUST 09:30 - 10:30 Details

MATTHIEU DESBRUN, Caltech University
10:30 am

Coffee break

KAUST 10:30 - 11:00 Details

11:00 am

On equilibrium shapes and fairness of polyhedral surfaces

Within the area of shape modeling in the presence of structural and fabrication constraints, we discuss three closely related topics: (i) 2D trusses with minimum weight (Michell structures), (ii) form-finding for architectural freeform shells and (iii) “kink-minimizing” polyhedral surfaces: All these themes are related to the minimization of total absolute curvature (integral of sum of absolute values of principal curvatures) of surfaces in various geometries and thus can be computationally treated with the same methodology. This is ongoing research with Martin Kilian, Davide Pellis and Johannes Wallner.

KAUST 11:00 - 11:30 Details

HELMUT POTTMANN, Technical University of Vienna
11:30 am

Connectivity Editing and Optimization for Polygonal Meshes

The connectivity of polygonal meshes is important in a wide range of applications. A very good example are architectural constructions that make the mesh lines and faces visible. I will talk about some past and present research involving the interactive connectivity editing of polygonal meshes and the automatic optimization of mesh connectivity. I will discuss examples from triangle meshes, quad meshes, quad dominant meshes, and mixed triangle / quad meshes.

  • PETER WONKA, KAUST

    PETER WONKA

KAUST 11:30 - 12:00 Details

PETER WONKA, KAUST
12:00 pm

Lunch / Prayer

KAUST Dining Hall – reserved section attendees only

KAUST 12:00 - 13:30 Details

1:30 pm

Fluid Simulations and Neural Networks

Physics simulations for virtual smoke, explosions or water are by now crucial tools for special effects. Despite their wide spread use, it is still difficult to get get these simulations under control, and they are still far too expensive for practical interactive applications. In this talk I will outline research directions to alleviate these inherent difficulties with machine learning techniques based on neural networks. These networks can learn and represent highly non-linear functions, which turns out to be highly useful in the flow simulation context.

A central part of this talk will be devoted to methods to enable the data-driven synthesis of fluid effects. In particular, I will outline methods to synthesize smoke volumes with pre-computed libraries of data, and a method the synthesizes liquid animations by deforming space-time data-sets of liquid surfaces. In both cases convolutional neural networks play a central role to make the approaches efficient and tractable.

I will show several examples of smoke and liquid animations generated with these approaches, and explain challenges encountered during data generation and training. In addition, I will discuss limitations and future directions of this new field of research.

KAUST 13:30 - 14:00 Details

NILS THUEREY, Technical University of Munich
2:00 pm

Balancing speed and fidelity in physics based animation

Physically based animation is a key part of many training simulations and entertainment applications. Given the unavoidable trade-off between high fidelity motion and interactive simulation rates, it is useful to have models and methods that find a good balance and provide a mechanism for selecting an appropriate level of fidelity. In this talk I will present work that explores this trade-off for the physically based simulation of contact, deformation, and articulated structures. I will describe examples such as numerical coarsening of elastic solids, discretization of contact at arbitrary resolutions, and reduced models for compliant structures.

KAUST 14:00 - 14:30 Details

PAUL KRY, McGill University
2:30 pm

On the Integration of Stiff Nonlinear Problems

We discuss a new integration algorithm for the accurate and efficient solution of stiff non-linear problems governed by the second-order ordinary differential equations, like the simulation of deformable bodies, textiles, and fibers. Traditional methods have the shortcoming that their performances are highly dependent on the numerical stiffness.
Advanced state-of-the-art methods in visual computing (like Gautschi-type exponential integrators) are most efficient, if the nonlinearity is moderately stiff. To overcome these limitations, we discuss a new integration method which is based on an exponential treatment of the full nonlinear forcing operator as opposed to more standard Gautschi-type exponential integrators, and the utilization of the concept of stiff accuracy. This results in significant increases of accuracy and efficiency, and allows for more complex and realistic models to be explored without compromising efficiency.

KAUST 14:30 - 15:00 Details

DOMINIK MICHELS, KAUST
3:00 pm

Coffee Break

KAUST 15:00 - 15:30 Details

3:30 pm

Faster PET Reconstruction with a Stochastic Primal-Dual Hybrid Gradient Method

In this talk we revisit the problem of Positron Emission Tomography (PET) reconstruction with non-smooth and convex priors. As the data fidelity term in PET is the Poisson likelihood, there are not many algorithms that can solve this problem. A very popular choice for solving it is the Primal-Dual Hybrid Gradient (PDHG) method proposed by Chambolle and Pock. While this algorithm works well for small to medium size problems in PET, in the case when the data size becomes large, computational issues arise, which is due to the fact that the system matrix for clinical PET scanners is very large and cannot be stored in the memory of most computers. Instead, an expensive algorithm to compute matrix-vector products has to be employed.

In this work we circumvent this issue by marrying the PDHG method with a randomized dual decomposition strategy (somewhat reminiscent of ART, Kaczmarz, and OSEM, which operate in the primal). PDHG arises as a special (and suboptimal) case of our more general method. We prove that our algorithm converges for a wide range of random samplings (e.g., uniform sampling, importance sampling, parallel samplings, ...), enabling us to implement various decomposition rules which can be chosen to fit the particular computing environment in use. We prove rigorous iteration complexity bounds for standard and accelerated variants of our method. Numerical examples show that the new method is much faster than PDHG.

This is joint work with Antonin Chambolle (Ecole Polytechnique), Matthias Ehrhardt (Cambridge), and Carola-Bibiane Schoenlieb (Cambridge).

KAUST 15:30 - 16:00 Details

PETER RICHTARIK, KAUST
4:00 pm

Topology-Varying Shape Matching and Modeling

The automatic creation of man-made 3D objects is an active area in computer graphics. Computer-assisted mixing and blending of components or subcomponents from existing example shapes can help users quickly produce interesting and creative designs. A key factor for automating this task is using algorithms that can match compatible parts between objects of different shape and structure. However, due to the coarse correspondence computed by current matching algorithms, automatic shape blending is mainly limited to the substitution of large compatible part sets.

We address the problem of relating 3D shapes of different geometry and topology, with applications in shape synthesis. Our goal is to compute a fine-grained mapping between two shapes differing in the geometry, cardinality, and connectivity of their parts, and to use this mapping for continuous shape interpolation.

First, we propose a framework for shape matching using a joint geometric and topological transformation. The framework follows the assumption that the best mapping for a pair of shapes is one that results from a shape transformation that minimally distorts the structural properties of a shape. We establish meaningful correspondences between shapes with large topological discrepancy by going beyond shape deformations and incorporating topological operations such as part split, duplication, and merging. We evaluate our correspondence algorithm on a diverse set of shape classes and compare the results to state-of-the-art methods.

Second, we propose an algorithm for synthesizing interpolations between structurally different 3D shapes. Our algorithm produces a continuous and plausible shape transformation that gradually morphs the geometry of the individual parts, as well as performs any necessary topology-changing operations. We further demonstrate the utility of our framework by developing intuitive shape creation tools. We show how these tools can allow novice users to synthesize new 3D models from continuous blends of topologically different shapes.

KAUST 16:00 - 16:30 Details

IBRAHEEM ALHASHIM, Simon Fraser University
4:30 pm

Algorithms for Geometrically-Structured Optimization

Many problems in geometry processing, graph theory, and machine learning involve optimizations whose variables are defined over a geometric domain. The geometry of the domain gives rise to geometric structure in the optimization problem itself. In this talk, I will show how leveraging geometric structure in the optimization problem gives rise to efficient and stable algorithms applicable to a variety of application domains. In particular, I will describe new methods for problems arising in shape analysis/correspondence, flows on graphs, and surface parameterization.

KAUST 16:30 - 17:00 Details

JUSTIN SOLOMON, MIT
9:15 am

Registration

KAUST 09:15 - 09:30 Details

9:30 am

Materials in the wild

Our daily lives bring us in contact with a rich range of materials that contribute to both the utility and aesthetics of our environment. Human beings are remarkably good at perceiving subtle distinctions in material appearance; e.g., is this fabric silk or cotton? is this surface granite or laminate?
But what makes silk look like silk? We are working on understanding why materials have their distinct appearance, and how humans perceive materials in their daily lives to understand the world they live in. This understanding has broad applications in vision and graphics in virtual and augmented reality, e-commerce and retail, and industrial and interior design.

KAUST 09:30 - 10:30 Details

KAVITA BALA, Cornell University
10:30 am

Poster Session

Hall 1

KAUST 10:30 - 00:00 Details

12:00 pm

Lunch / Prayer

KAUST Dining Hall – reserved section attendees only

KAUST 12:00 - 14:00 Details

2:00 pm

Data-Driven Interpolation of Optical Material Properties

Data-driven material representations enable the faithful reproduction of real world material on novel geometries and under arbitrary viewing and lighting conditions in synthetic images. One such representation is the Bidirectional Texture Function (BTF), which can be acquired from a material sample via a measurement process that takes a large number of images from multiple view points and under multiple illumination directions. This type of representation is well suited to accurately reproduce a given material sample, but editing a measured material or even designing a new one based on multiple measurements is a difficult task for an artist. This is due to the high complexity and size of the data-set and the necessity to edit all views in a consistent way. One editing paradigm that can be used for this is interpolation, which enables the creation of a novel material that lies perceptually in between several measured samples. This talk will describe a texture-synthesis based interpolation technique that can be used to create believable interpolated materials and also continuous interpolation sequences. It can be used even for materials with complex feature topology, spatially varying reflectance behavior and a meso-structure causing strong parallaxes in the measured images.

KAUST 14:00 - 14:30 Details

ROLAND RUITERS, Autodesk
2:30 pm

Computational Modeling of the Growth of Plants

Plant growth is a complex, dynamic pattern for which the underlying biological mechanisms are still unknown. In this talk, a mathematical model of 3D tree growth is discussed that emphasizes the impact of self-organization of branches in space on tree form. The model's relatively low-dimensional parameter space captures a large diversity of realistic tree architectures.

KAUST 14:30 - 15:00 Details

WOJCIECH PALUBICKI, Adam Mickiewicz University
3:00 pm

Visual Computing Challenges at Autodesk

In the past, Autodesk’s computer graphics research was mainly driven by classical visualization needs for visual effects, product design and construction.
While visualization continues to be an important area of research, particularly in the context of virtual prototyping, other topics related to multi-user workflows, artificial intelligence as well as data analytics and exploration have gained importance.
In this presentation I will discuss Autodesk’s visual computing challenges and how fundamental research can help to solve some of the underlying problems

KAUST 15:00 - 15:30 Details

ARNO ZINKE, Autodesk
3:30 pm

Coffee Break

KAUST 15:30 - 16:00 Details

4:00 pm

Simulation Enabled Discoveries: Case Studies in MHD and Fluid Dynamics

Computation has emerged as the indispensible third leg of scientific discovery along with the traditional two branches of theory and experiment. In this talk we discuss some case studies from magnetohydrodynamics (MHD) and fluid dynamics based on research conducted in the Fluid & Plasma Simulation Laboratory.
Our first case study is that of magnetic suppression of the Richtmyer-Meshkov instability (RMI). In hydrodynamics, a density interface separating two fluids is unstable under impulsive acceleration. This instability aka RMI has been the bane of inertial confinement fusion where it is considered highly detrimental. The suppression of this instability in the presence of a magnetic field was a purely serendipitous discovery due to large-scale simulations.
The second case study is that of magnetic reconnection (MR). Loosely defined, MR is the breaking and reattachment of magnetic field lines, and is a somewhat ubiquitous phenomenon in our universe: it accompanies coronal mass ejections from our Sun, observed in the earth’s magneto-tail, etc. Standard quasi-steady models in single-fluid resistive MHD severely under predict the reconnection rate compared with observations. A key-missing piece of physics is an instability dubbed the plasmoid instability. Large-scale MHD simulations led to the discovery of this instability demonstrating that the reconnection rate saturates for Lundquist number greater than 104.
The fluid dynamics examples pertain to simulations of wall-bounded turbulence. Here we focus on the simulation of the drag crisis in flow over bluff bodies and show how large-eddy simulations can successfully reproduce this phenomenon.
We believe these case studies amply demonstrate that simulation enabled discoveries are no longer a myth, and that computations are a legitimate tool to answer clearly posed questions in science and engineering

  • RAVI SAMTANEY, KAUST

    RAVI SAMTANEY

KAUST 16:00 - 16:30 Details

RAVI SAMTANEY, KAUST
4:30 pm

Computational Design of Deforming Objects for Advanced Fabrication

I will describe recent progress in the area of computational fabrication towards novel concepts for reproducing objects with nontrivial shapes and topologies. First, I will present FlexMolds, a novel computational approach to automatically design flexible, reusable molds that, once 3D printed, allow us to physically fabricate, by means of liquid casting, multiple copies of complex shapes with rich surface details and complex topology. I will then investigate the design of objects that can self-deform. I will introduce CurveUps, curvy shells that form from an initially flat state. They consist of small rigid tiles that are tightly held together by two pre-stretched elastic sheets attached to them. Our method allows the realization of smooth, doubly curved surfaces that can be fabricated as a flat piece. Once released, the restoring forces of the pre-stretched sheets support the object to take shape in 3D. CurveUps are structurally stable in their target configuration. All approaches will be illustrated with examples.

KAUST 16:30 - 17:00 Details

BERND BICKEL, Institute of Science and Technology Austria
7:00 pm

Conference Gala dinner at Al Marsa Restaurant

(attendees only)
Poster session awards

KAUST 19:00 - 21:30 Details

8:45 am

Registration

KAUST 08:45 - 09:00 Details

9:00 am

Fluid Imaging for Graphics and Beyond

Fluid imaging has many significant applications in scientific and engineering fields such as combustion research, design of airplanes and underwater vehicles, and development of artificial heart valves. It has also been intensely studied in computer graphics. Since 3D unsteady flows and turbulence are very common in such domains, the main task of the fluid imaging is to allow probing the fluid motions over a range of length scales. In other words, the ultimate goal is to be able to obtain 3D dense measurements of the three components of the velocity vector, known as 3D-3C. Unfortunately, a dense reconstruction of all three components of the velocity field over the full 3D volume requires multiple cameras and remains elusive in practice. In this presentation I will summarize recent work at KAUST to develop a single-camera, full 3D fluid capture method.

KAUST 09:00 - 09:30 Details

WOLFGANG HEIDRICH, Director VCC
9:00 am

Industrial Symposium: Registration and Networking

KAUST 09:00 - 10:00 Details

9:30 am

Fast Transient Imaging with AMCW Lidar Systems

AMCW lidar systems such as the ones used in Microsoft Kinect for Xbox One, Microsoft HoloLens or Google Tango smartphones provide affordable full-field range imaging at high frame rates. They rely on indirect measurements of the time of flight of light. However, the assumption of a single time of flight per pixel is inadequate in reality. Global illumination effects lead to light paths of varying length which interfere and lead to strong, systematic distortions in range images. To handle this problem correctly, a full reconstruction of the time-dependent impulse response of light is needed per pixel. Such images of impulse responses are known as transient images. Reconstructing them from AMCW lidar measurements is a non-trivial inverse problem. We demonstrate that a specific measurement procedure turns this problem into a trigonometric moment problem. These problems are well-studied and powerful closed-form solutions exist. The resulting techniques serve to reconstruct complex impulse responses from measurements at few (e.g. three) modulation frequencies. The approach scales well from interactive capture of range images with reduced multipath interference to capture of high-quality transient images. Good transient images enable a more complete understanding of light-transport in the captured scene. The research further opens up new ways to analyze the reconstruction problem itself. Most of the presented work has been published at SIGGRAPH Asia 2015.

KAUST 09:30 - 10:00 Details

CHRISTOPH PETERS, University of Bonn
10:00 am

Coffee break

KAUST 10:00 - 10:30 Details

10:00 am

Industrial Symposium: Opening

Symposium opening
Mr. Tristan Walker, VP, Innovation and Economic Development
VCC Overview, Prof. Wolfgang Heidrich, VCC Director

KAUST 10:00 - 10:30 Details

10:30 am

The Neuron Reconstruction Algorithm

Modern microscopic techniques enable imaging of three-dimensional neuron morphologies in intact brain tissue. In principle it is now possible to automatically reconstruct the dendritic branching patterns of neurons from 3D fluorescence image stacks. In practice however, the signal-to-noise ratio can be low, in particular in the case of thin dendrites or axons imaged relatively deep in the tissue. Here we present a nonlinear anisotropic diffusion filter that enhances the signal-to-noise ratio while preserving the original dimensions of the structural elements. The key idea is to use structural information in the raw data — the local moments of inertia — to locally control the strength and direction of diffusion filtering. A cylindrical dendrite, for example, is effectively smoothed only parallel to its longitudinal axis, not perpendicular to it. The filter is a valuable general tool for smoothing cellular processes and is well suited for preparing data for subsequent image segmentation and cell reconstruction.

References:
P. J. Broser, R. Schulte, A. Roth, F. Helmchen, J. Waters, S. Lang, B. Sakmann, G. Wittum: Nonlinear anisotropic diffusion filtering of three-dimensional image data from 2-photon microscopy. Journal of Biomedical Optics 9(6), 1253–1264 (2004)
Queisser, G., H. Bading, M. Wittmann, G. Wittum: Filtering, reconstruction and measurement of the geometry of neuron cell nuclei based on confocal microscopy data, Journal of Biomedical Optics, Jan 2008.
Daniel Jungblut, Andreas Vlachos, Gerlind Schuldt, Nadine Zahn, Thomas Deller, Gabriel Wittum: SpineLab – a Tool for Three Dimensional Cell Morphology Reconstruction. J Biomed Opt., 17(7), (2012)

  • GABRIEL WITTUM, KAUST

    GABRIEL WITTUM

KAUST 10:30 - 11:00 Details

GABRIEL WITTUM, KAUST
10:30 am

Industrial Symposium: Session I

Scanning, Reconstruction, and Fabrication of 3D Objects
Peter Wonka
Survey-Grade Photogrametry and Modeling
Neil Smith

  • PETER WONKA, KAUST

    PETER WONKA

KAUST 10:30 - 11:15 Details

PETER WONKA, KAUST
11:00 am

Some structural properties of functional map computation

In recent years, several techniques have been proposed for computing and manipulating correspondences between shapes using the so-called functional map framework. Although capable of producing high quality results for problems such as shape matching and tangent vector field design, the exact structural properties, and in particular, the settings in which it can lead to point-to-point maps are not well understood. In this talk, I will give a brief overview of the basic ideas behind the functional map framework and then describe some recent work that allows to both incorporate and extract additional information from functional maps. Namely, I will discuss ways to characterize deformation fields as linear operators and a technique to use more information from a given set of descriptor constraints to obtain functional maps that follow structural properties of pointwise correspondences

KAUST 11:00 - 11:30 Details

MAKS OVSJANIKOV, Ecole Polytechnique
11:15 am

Industrial Symposium: Coffee Break

KAUST 11:15 - 11:30 Details

11:30 am

From Real-Time Shape Deformation to Geometric Flows in Shape Space

The optimization of deformable, flexible or non-rigid shapes is essential for many tasks in geometric modeling and processing. In the first part of this talk, I will introduce model reduction techniques that can be used to construct fast approximation algorithms for shape optimization problems. The goal is to obtain run times that are independent of the resolution of the discrete shapes to be optimized. As an example, we will discuss a method for real-time elasticity-based shape interpolation.

In the second part, we will broaden the perspective and discuss how concepts from elasticity can be used to obtain geometric structures on shape spaces, in which a shape is a single point. We will see how these structures can be used for the processing of motion and animations of non-rigid shapes. The idea is to treat the motions as curves in shape space and to transfer concepts from curve processing in Euclidean space to the processing of motion of non-rigid shapes. We will discuss explicit examples including a geometric flow of curves in shapes space that can be used for reducing jittering artifacts in motion capture data, the construction of subdivision curves in shape space and the efficient computation of geodesics in shape space.

KAUST 11:30 - 12:00 Details

KLAUS HILDEBRANDT, Technical University of Delft
11:30 am

Industrial Symposium: Session II

On the Simulation of Nonwoven Fabric Manufacturing
Dominik Michels
TBA
Bernard Ghanem

KAUST 11:30 - 12:15 Details

DOMINIK MICHELS, KAUST
12:00 pm

VCC Open House, Exhibition and Lunch

Demos and Exhibitions from the Visual Computing Center (VCC),
the KAUST Visualization Lab (KVL) and local industry

A light lunch will be served at the center

“Sponsored by the KAUST Industry Collaboration Program (KICP)”

KAUST 12:00 - 14:00 Details

2:00 pm

Industrial Symposium: Explore Kaust

KAUST Museum and King Abdullah Monument

KAUST 14:00 - 16:00 Details

4:00 pm

Canceled because of the weather: [South Beach - Social]

KAUST 16:00 - 18:00 Details

6:30 pm

Dinner- Al-Marsa Yacht Club Restaurant

Buses will move from KAUST Inn between 06:00 and 06:15 PM
Al-Marsa Yacht Club Restaurant

KAUST 18:30 - 20:00 Details