Invited talks

Leonidas Guibas
Stanford University


Networks of Shapes and Images


Robotic agents operating autonomously in the world need an understanding of their environment that goes beyond what can be literally inferred from their sensor data. Successful operation requires a contextual understanding of the geometry and physics of the objects and other actors present in a scene that is based on prior knowledge, analogical reasoning, and short- and long-term memory (in the robot or the cloud). To make this possible we need tools establish semantic relationships between objects models and their sensor equivalents which allow information and knowledge transfer. The goal of this talk is to present a set of tools for linking images, videos, 3D scans and other sensory modalities to each other and to 3D models, forming networks that enable useful information transport.

A key tool in this process is the representation of correspondences or maps between related data within a single modality and across different modalities. We aim to make relationships or correspondences between data sets first-class citizens -- so that the relationships themselves become explicit, algebraic, storable and searchable objects. Networks of such relations can interconnect data sets into societies where the "wisdom of the collection" can be exploited in performing operations on individual data sets better, or in further assessing relationships between them. Examples include entity extraction from images or videos, 2D or 3D segmentation, the propagation of annotations and labels among images/videos/3D models, variability analysis in a collection of shapes, etc.

The talk will cover general mathematical and computational tools for the construction, analysis, and exploitation of such relational networks -- illustrated by several concrete examples using 3D models and/or 2D images. By creating societies of data sets and their associations in a globally consistent way, we enable a certain joint understanding of the data that provides the powers of abstraction, analogy, compression, error correction, and summarization. Furthermore, given live sensor data, this machinery allows the inference of additional structure not present in the original signal, such as depth estimation in images, the geometry of occluded areas, etc.

This "functorial" view of geometric data puts the spotlight on consistent, shared relations and maps as the key to understanding structure and making inferences. It is a little different from the current dominant paradigm of extracting supervised or unsupervised feature sets, defining distance or similarity metrics, and doing regression or classification - though representation sparsity still plays an important role. The inspiration is more from ideas in functional analysis and homological algebra, exploiting the algebraic structure of data relationships or maps in an effort to disentangle dependencies and assign importance to the vast web of all possible relationships among multiple geometric data sets.


Professor Guibas heads the Geometric Computation group in the Computer Science Department of Stanford University and is a member of the Computer Graphics and Artificial Intelligence Laboratories. He works on algorithms for sensing, modeling, reasoning, rendering, and acting on the physical world. Professor Guibas' interests span computational geometry, geometric modeling, computer graphics, computer vision, sensor networks, robotics, and discrete algorithms - all areas in which he has published and lectured extensively. Examples of current and recent activities include: data structures for mobile data (kinetic data structures), ad-hoc sensor and communication networks, randomized geometric algorithms, rounding and approximating geometric structures, local and global analysis with point cloud data, Monte-Carlo algorithms for global illumination and motion planning, organizing and searching libraries of 3D shapes and images, physical simulations involving deformations and contacts, estimation of mappings between 3D shapes, intelinking image collections, analysis of GPS traces and other mobility data. Leonidas Guibas obtained his Ph.D. from Stanford in 1976, under the supervision of Donald Knuth. His main subsequent employers were Xerox PARC, MIT, and DEC/SRC. He has been at Stanford since 1984 as Professor of Computer Science. He has produced several Ph.D. students who are wellknown in computational geometry, such as John Hershberger, Jack Snoeyink, and Jorge Stolfi, or in computer graphics, such as David Salesin, Eric Veach, or Niloy Mitra. At Stanford he has developed new courses in algorithms and data structures, geometric modeling, geometric algorithms, and sensor networks. Professor Guibas is an ACM and IEEE Fellow as well as a winner of the ACM/AAAI Allen Newell award.