Sponsor:

*VIRTUAL* – October 18th-22nd, 2021

Co-located with SVR 2021 e SBGames 2021

SIBGRAPI 2021 Preliminary Program

The program listed below is preliminary and subject to change without warning. Please do not use this program for planning your participation in the event.

A more detailed and final program will be made available in this page as we get closer to the event date.

Click on the image below to open it in a new window in full size.

Times specified in Brasilia Time (BRT), UTC -3.

SIBGRAPI 2021 Keynote Speakers

Photo of Cláudio Rosito Jung.

Cláudio Rosito Jung
UFRGS, Brazil

Talk Title: Beyond Horizontal Bounding Boxes for Object Detection

Talk Abstract: Object Detection is a classical problem in computer vision, and it consists of identifying each instance of a set of categories in a given image. Representing objects as horizontal bounding boxes (HBBs) allows easy annotation and fewer parameters to regress, but might be a poor fit for rotated or deformable objects. On the other hand, full segmentation masks provide a complete representation of the object, but annotation is tedious and regression typically involves a non-parametric representation. In this talk I will discuss a few alternatives inbetween these two extrema: first I will talk about warped planar representations (WPR), in which the object is represented as an affine-transformed HBB; then I will talk about a fuzzy representation based on Gaussian Bounding Boxes (GBBs), which can be mapped either to rotated ellipses or Oriented Bounding Boxes (OBBs) and allows the use of a differentiable loss function that resembles the Intersection-over-Union (IoU).

Short Bio: Cláudio Rosito Jung received the B.S. and M.S. degrees in Applied Mathematics, and the Ph.D. degree in Computer Sciences, from Universidade Federal do Rio Grande do Sul (UFRGS), Brazil, in 1993,1995 and 2002, respectively. He is currently a faculty member at UFRGS in the Computer Science department, and was a visiting faculty at the University of Pennsylvania from july 2015 to july 2016. His research interests include several aspects in image processing, computer vision, pattern recognition and deep learning, such as biomedical imaging, multiscale image analysis, intelligent vehicles, multimedia applications, human motion synthesis and analysis, audiovisual signal processing, stereo/multiview matching and spherical image processing.

Photo of Gladimir Baranoski.

Gladimir Baranoski
University of Waterloo, Canada

Talk Title: The Quest for Fundamental Biophysical Data

Talk Abstract: Predictive models of light and matter interactions are employed in a wide range of applications in several fields from computer graphics and remote sensing to biomedical optics and photonics, just to name a few. It is a well-known fact that a well-designed model is of little use without reliable specimen characterization data (e.g., thickness and pigment concentrations) to be used as input, and reliable evaluation data (e.g., spectral reflectance and transmittance) to be used in the assessment of its predictive capabilities. Ideally, the specimen’s characterization data to be incorporated into a model should correspond to the specimen used to obtain the measured data employed in its evaluation. However, the few spectral datasets available in the literature rarely provide a comprehensive description of the target specimens. Data is even more scarce for materials in their pure form, such as natural pigments, whose absorption profile is often obtained either through inversion procedures, which may be biased by the inaccuracies of the inverted model, or does not take into account in vivo and in vitro discrepancies. In this talk, we informally address these issues and their practical implications for the development of robust hyperspectral technologies relying on light interaction models.

Short Bio: Gladimir V. G. Baranoski received his doctoral degree in computer science from the University of Calgary (Canada) in 1998. He is currently a Professor of the School of Computer Science at the University of Waterloo (Canada), where he has established the Natural Phenomena Simulation Group (NPSG). His research interests include primarily the predictive simulation of light interactions with natural materials. As the leader of NPSG, he has been actively participating in the development of a wide range of hyperspectral light interaction models aimed at interdisciplinary investigations. The results of his research have been disseminated in well- known venues of different fields, including computer graphics, remote sensing, and biomedical optics. He has also organized conference courses and published books related to his research work.

Photo of Hugo Proença.

Hugo Proença
University of Beira Interior, Portugal

Talk Title: Advances in Visually Interpretable Biometric Recognition

Talk Abstract: There is a popular adage that states that “a picture is worth more than 1,000 words”. Complex ideas are known to be conveyed in a more effective way by a single still image than by a verbal description. Among the many examples in the literature supporting this idea, we highlight Leonardo da Vinci’s, who wrote that a poet would be "overcome by sleep and hunger before being able to describe with words what a painter is able to depict in an instant" or Napoleon Bonaparte, who is supposed to have said that “un bon croquis vaut mieux qu'un long discours".
  In this talk, we will discuss the development of methods to simultaneously (and jointly) “recognize” objects in images and “interpret” their decisions. In particular, we consider the biometric (periocular) recognition problem. In practice, the idea is to obtain classification models that not only infer class (ID) information , but also provide local and global visualizations of the data that justified the decision: e.g., by providing synthetic representations of the input data, that - in a human understandable way - justify why a pair of samples is/is not from the same person.
  Such “visually interpretable recognizers” have various applications, such as: 1) support human-decision processes in domains such as forensics, where the results of one analysis should be communicated in a human-understandable way to a jury or a judge; 2) improve biometrics feature engineering, by providing cues about the features that are of most interest to provide good decisions; 3) drive future data collection protocols, by providing a good understanding of the value of each data segment privileged by the recognition system; and 4) augment trust, by showing insights that fit the usual human understanding of the recognition problem, even for people with no knowledge in data science.

Short Bio: Hugo Proença (SM'12), B.Sc. (2001), M.Sc. (2004) and Ph.D. (2007) is an Associate Professor in the Department of Computer Science, University of Beira Interior and has been researching mainly about biometrics and visual-surveillance. He was the coordinating editor of the IEEE Biometrics Council Newsletter and the area editor (ocular biometrics) of the IEEE Biometrics Compendium Journal. He is a member of the Editorial Boards of the Image and Vision Computing, IEEE Access and International Journal of Biometrics. Also, he served as Guest Editor of special issues of the Pattern Recognition Letters, Image and Vision Computing and Signal, Image and Video Processing journals.

Photo of Luis Gustavo Nonato.

Luis Gustavo Nonato
ICMC-USP, Brazil

Talk Title: From Geometry Processing to Spatio-Temporal Data Analysis and Visualization: a historical view

Talk Abstract: In this talk, we will revisit the work carried out by the research group led by prof. Nonato and his collaborators in the last decades. From a historical perspective, we will show the group's contributions in the areas of geometric processing, visualization, and spatio-temporal visual data analysis, overviewing the mathematical and computational tools employed to tackle problems ranging from mesh generation from images to crime pattern analysis.

Short Bio: Luis Gustavo Nonato received the PhD degree in applied mathematics from the Pontifícia Universidade Católica do Rio de Janeiro, Rio de Janeiro - Brazil, in 1998. He is professor in the Instituto de Ciências Matemáticas e de Computação, Universidade de São Paulo, São Carlos, Brazil. Nonato was a visiting professor at the Center for Data Science, New York University, New York - USA from 2016 to 2018 and he was also a visiting scholar in the Scientific Computing and Imaging Institute, University of Utah, Salt Lake City - USA from 2008 to 2010. Besides having served in several program committees, including IEEE SciVis, IEEE InfoVis, and EuroVis, Nonato was associate editor of the Computer Graphics Forum journal and currently he is associate editor of the IEEE Transactions on Visualization and Computer Graphics. He is also editor-in-chief of the SBMAC SpringerBriefs in Applied Mathematics and Computational Sciences. Nonato's main research interests include visual analytics, geometric computing, data science, and visualization. Nonato has a strong interest in bridging the gap between academia, industry, and governments, leading a number of initiatives with the private sector and government agencies.

SIBGRAPI 2021 WIA Invited Talks

Photo of Marcelo Siqueira.

Marcelo Siqueira
Align Technology, Inc.

Talk Title: From trimmed NURBS to watertight boundary representations of CAD models

Talk Abstract: A fundamental problem in Geometric Modeling is the one of converting the representation of a solid from one (representation) form to another. In this talk, I will discuss a particular instance of this problem, namely, the conversion of boundary representations of CAD models from the trimmed Non-Uniform Rational B-Splines (NURBS) form to watertight parametric forms. For historical and for practical reasons, the geometry of CAD models have been represented by trimmed NURBS. While this representation form is often more convenient to control a signature piece of the boundary in isolation, it is in general very difficult or impossible to represent the entire boundary as a watertight surface. In other words, the trimmed NURBS surface patches may not join each other continuously, giving rise to a boundary representation with very small gaps (i.e., fillets) along the boundary curves of the trimmed patches. With the advent and recent progress of Iso-Geometric Analysis (IGA), the need for watertight boundary representations of CAD models prompted several researchers to pay close attention to the conversion problem again. The aim of my talk is two-fold. First, I will focus on the shortcomings of the most recent solutions to the conversion problem from an industry's perspective. Second, I will shed some light on what constitutes a good solution to the problem. The talk is related to an on-going research collaboration with Jörg Peters (UF, United States) and Paulo Pagliosa (UFMS, Brazil).

Short Bio: Marcelo Siqueira received the BSc (computer science), in 1992, from Universidade Federal do Rio Grande do Norte (UFRN). He received the master degree, in 1994, from Universidade de SĂŁo Paulo (USP), and the PhD degree, in 2006, from the University of Pennsylvania (UPenn). From 1996 to 2008, he was a professor at the College of Computing of Universidade Federal de Mato Grosso do Sul (UFMS). From 2009 to 2017, he was a professor at the Department of Mathematics of UFRN. In 2017, he joined Velo3D, a startup located at Campbell, CA, USA, that developed a disruptive technology for 3d printing of metallic parts. In 2018, he joined Align Technology, the world leader of clear aligners for orthodontic therapy, where is currently a full-time researcher and software developer. His research interests are mesh generation, modeling of curves and surfaces, and digital topology.

Photo of Vanessa Testoni.

Vanessa Testoni
Samsung Research Brazil

Talk Title: Available soon

Talk Abstract: Available soon

Short Bio: Vanessa Testoni is the Multimedia team leader at SRBR (Samsung Research Brazil), where her research interests are a mix of image/video coding, processing and streaming, video standards, machine learning, computer vision and information theory. She received her BS in CS from PUCPR (Brazil), her EE degree from UFPR (Brazil) and her MSc and PhD degrees in EE from UNICAMP (Brazil). She started her career at Siemens Telecommunications and later joined UCSD (USA) as a postdoctoral employee. She was awarded the Microsoft Research PhD Fellowship Award and the MIT TR35 (Young Innovators under 35) in the first Brazilian edition of the award. She is an IEEE senior member, an elected affiliate member of the Brazilian Academy of Sciences (ABC), a previous chair of the IEEE SPS (Signal Processing Society) SĂŁo Paulo chapter and was nominated the first national head of the ISO/IEC JTC 001/SC 29 (JPEG/MPEG) Brazilian delegation.

SIBGRAPI 2021 WVIS Invited Talk

Photo of Fernando Paulovich.

Fernando Paulovich
Dalhousie University, Canada

Talk Title: From Visual Analytics to Explainable AI: the Ingredients of More Reliable Classification Models

Talk Abstract: In recent decades, classification models have proven to be essential machine learning tools due to their potential and applicability in various domains. In these years, the general direction of most researchers has been to improve quantitative metrics, despite the lack of information about models' inner workings such metrics convey. This paradigm is shifting, and strategies beyond tables and numbers to help interpret model decisions are gaining importance. As part of this trend, visualization and visual analytics tools and techniques have been widely used and shown to be essential ingredients for implementing the so-called explainable Artificial Intelligence (XAI) concept. In this talk, I will introduce the idea of visual analytics and discuss how it has been used to interpret classification models, support the understanding of models' general behavior, and audit the produced results to increase confidence in the predictive analytics process.

Short Bio: Fernando V. Paulovich is an associate professor and Canada Research Chair in Data Visualization at the Faculty of Computer Science, Dalhousie University, and head of the Visualization and Visual Analytics (VVA) lab. Over the past ten years, he has been researching in the field of computational visualization, more specifically information visualization, visual analytics and visual data mining. His focus is on integrating machine learning and visualization tools and techniques, taking advantage of the “intelligence” provided by machine learning approaches, and of user knowledge by means of interactions with visual representations, helping people to understand and take full advantage of this "brave new information world."

Organized by

Sponsored by

Support by

In-Cooperation with