MONDAY, MAY 6th 2019

09:00 - 17:00 Room: MAESTRALE	T1	Deep Learning for Computer Graphics and Geometry Processing
09:00 - 12:30 Room: PONENTE	T2	libigl: Prototyping Geometry Processing Research in C++
09:00 - 17:00 Room: TRAMONTANA	T3	Smoothed Particle Hydrodynamics for Physics-Based Simulation of Fluids and Solids
13:30 - 17:00 Room: PONENTE	T4	Learning Generative Models of 3D Structures

---

 

T1 Deep Learning for Computer Graphics and Geometry Processing

Bronstein, Michael (Imperial College London/USI Lugano); Guibas, Leonidas (Stanford/Facebook); Kokkinos, Iasonas (University College London); Litany, Or (Stanford/Facebook); Mitra, Niloy (University College London); Monti, Federico (USI Lugano); Rodolà, Emanuele (La Sapienza).

Time & Location: 09:00 - 17:00 Room: MAESTRALE
Duration: Full day
Abstract:
In computer graphics and geometry processing, many traditional problems are now becoming increasingly handled by data-driven methods. In an increasing variety of problem settings, deep networks are state-of-the-art, beating dedicated hand-crafted methods by significant margins. This tutorial gives an organized overview of core theory, practice, and graphics-related applications of deep learning.

---

 

T2 libigl: Prototyping Geometry Processing Research in C++

Jacobson, Alec (University of Toronto); Panozzo, Daniele (NYU)

Time & Location: 09:00 - 12:30 Room: PONENTE
Duration: Half day
Abstract:
Modern geometry processing algorithms depend on an ever-growing toolbox of fundamental sub-routines and data structures. Prototyping from scratch requires much time building basic tools rather than focusing on the novel research idea. Many existing code libraries have unsatisfactory APIs and the time spent implementing sub-routines is often replaced with time spent learning complex, templated object hierarchies or memory layouts. Libigl is a C++ library of geometry processing algorithms designed for and by researchers. Its wide functionality includes construction of common sparse discrete differential geometry operators (such as the cotangent Laplacian), simple facet- and edge-based topology data structures, mesh-viewing utilities for OpenGL and GLSL, and many core functions for matrix manipulation which make Eigen feel a lot more like MATLAB. Libigl places extreme importance on ease of use and experimentation. To this end, algorithms are directly exposed as functions taking simple matrix types as inputs and outputs. Libigl is a "header only"library and compiles on Windows, Mac, and Linux. In this course, we will walk through an introduction of libigl via readymade examples spanning the gamut of geometry processing applications and tasks. Attendees will be able to follow along on their laptops. We will explain the core functionality of libigl, how to piece together complex algorithms from library functions, and how to interface to libigl from Python and MATLAB. We will highlight some of libigl's most powerul features: including mesh booleans, quad remeshing, parameterization, and shape deformation. We will conclude with live coding sessions demonstrating libigl's effectiveness and ease-of-use. The course continues beyond the lecture via libigl's interactive online tutorial complete with over 50 example demos (http://libigl.github.io/libigl/) and two open source graduate-level courses on geometry processing based on libigl (https://github.com/alecjacobson/geometry-processing) and (https://github.com/danielepanozzo/gp).

---

 

T3 Smoothed Particle Hydrodynamics for Physics-Based Simulation of Fluids and Solids

Koschier, Dan (UCL); Bender, Jan (RWTH Aachen); Solenthaler, Barbara (ETH); Teschner, Matthias (Univ.Freiburg);

Time & Location: 09:00 - 17:00 Room: TRAMONTANA
Duration: Full day
Abstract:
Graphics research on Smoothed Particle Hydrodynamics (SPH) has produced incredible and unique visual results that cannot be found in any other research community concerned with SPH simulations. Generally, the SPH formalism serves as a spatial discretization technique, commonly used for the numerical simulation of continuum mechanical problems such as the simulation of fluids, granular materials, and deformable solids. Recent advances in the field have made it possible to efficiently simulate massive scenes with highly complex boundary geometries on a single PC. Moreover, novel techniques allow to robustly handle interactions among various materials. As of today, graphics-inspired pressure solvers, neighborhood search algorithms, boundary formulations, and other contributions often serve as core components in commercial software for animation purposes as well as in computer-aided engineering software. The proposed tutorial will cover various aspects of SPH simulations. Governing equations for mechanical phenomena and their SPH discretizations will be discussed. Concepts and implementations of core components such as neighborhood search algorithms, pressure solvers, and boundary handling techniques will be presented. Implementation hints for the realization of SPH solvers for fluids, granular materials, elastic solids, and rigid bodies will be given. The tutorial aims to combine the introduction of theoretical concepts with the presentation of actual implementations. All presenters are experienced researchers, teachers, and practitioners in the area of SPH.

---

 

T4 Learning Generative Models of 3D Structures

Chaudhuri, Siddhartha (IIT Bombay); Xu, Kai (National Univ. of Defense Technology); Ritchie, Daniel (Brown Univ.); Zhang, Hao (Richard) (Simon Fraser Univ.)

Time & Location: 13:30 - 17:00 Room: PONENTE
Duration: Half day
Abstract:
In this tutorial, we give a full account of important techniques and directions for building generative models of 3D structures. The first half of the tutorial will be introductory, providing both a broad overview of the field as well as a quick refresher of important algorithmic ideas from geometric analysis and machine learning. The second half will consist of a deep dive into the most exciting methods for building generative models of single shapes and composite scenes. We will highlight how standard data-driven methods need to be adapted, or new methods developed, in order to create models that are both generative and structure-aware. At the end, attendees should come away with a historical context, a high-level understanding of all relevant work in the area, and familiarity with the mathematical tools to explore further.