April 22, 2024

Eurographics, the premier conference on computer graphics, will be held in Limassol, Cyprus this year (April 22-26, 2024). GrUVi lab will once again have a good show at Eurographics 2024, with:

• Han-Hung Lee (along with Manolis Savva and Angel X. Chang) presenting Text-to-3D Shape Generation STAR.
• Zhiqin Chen receiving 2024 Eurographics PhD Award for Best PhD Thesis.
• Ali Mahdavi-Amiri serving as the International Program Committee member.

Congratulations!

March 18, 2024

3D Vision (3DV), the premier conference on computer vision, will be held in-person in Davos, Switzerland on March 18-21, 2024. GrUVi lab will once again have a good show at 3DV 2024, with 2 technical papers. And Andrea Tagliasacchi will be serving as the program co-chair for the conference.

Congratulations to all the authors! And here are the 2 accepted papers:

Generalizing Single-View 3D Shape Retrieval to Occlusions and Unseen Objects

OPDMulti: Openable Part Detection for Multiple Objects

February 23, 2024

VCR seminar will hold a workshop for AAAI visitors from other universities. Below is the schedule for the talks.

• 11 am-11:40 am Levi Lelis (U of Alberta)
• 11:40 am-12:20 pm Vahid Babaei (MPI) Inverse Design with Neural Surrogate Models
• 12:20 pm -1:00 pm lunch break
• 1 pm -1:40 pm Sven Koenig (USC) Multi-Agent Path Finding and Its Applications
• 1:40 pm -2:20 pm Jiaoyang Li (CMU) Layout Design for Large-Scale Multi-Robot Coordination

Talk 1 (11 am-11:40) Levi Lelis, Department of Computing Science, University of Alberta

Title: Learning Options by Extracting Programs from Neural Networks

Abstract: In this talk, I argue for a programmatic mindset in reinforcement learning, proposing that agents should generate libraries of programs encoding reusable behaviors. When faced with a new task, the agent learns how to combine existing programs and generate new ones. This approach can be helpful even when policies are encoded in seemingly non-decomposable representations like neural networks. I will show that neural networks with piecewise linear activation functions can be mapped to a program with if-then-else structures. Such a program can then be easily decomposed into sub-programs with the same input type of the original network. In the case of networks encoding policies, each sub-program can be seen as an option---a temporally extended action. All these sub-programs form a library of agent behaviors that can be reused later, in downstream tasks. Considering that even small networks can encode a large number of sub-programs, we select sub-programs that are likely to generalize to unseen tasks. This is achieved through a subset selection procedure that minimizes the Levin loss. Empirical evidence from challenging exploration scenarios in two grid-world domains demonstrates that our methodology can extract helpful programs, thus speeding up the learning process in tasks that are similar and yet distinct from the one used to train the original model.

Bio: Dr. Levi Lelis is an Assistant Professor at the University of Alberta, an Amii Fellow, and a CIFAR AI Chair. Levi’s research is dedicated to the development of principled algorithms to solve combinatorial search problems. These problems are integral to optimizing tasks in various sectors. Levi’s research group is focused on combinatorial search problems arising from the search for programmatic solutions---computer programs written in a domain-specific language encoding problem solutions. Levi believes that the most promising path to creating agents that learn continually, efficiently, and safely is to represent the agents’ knowledge programmatically. While programmatic representations offer many advantages, including modularity and reusability, they present a significant challenge: the need to search over large, non-differentiable spaces not suited for gradient descent methods. Addressing this challenge is the current focus of Levi’s work.

Talk 2 (11:40 am-12:20) Vahid Babaei , Max Planck Institute for Informatics

Title: Inverse Design with Neural Surrogate Models

Abstract: The digitalization of manufacturing is turning fabrication hardware into computers. As traditional tools, such as computer aided design, manufacturing, and engineering (CAD/CAM/CAE) lag behind this new paradigm, the field of computational fabrication has recently emerged from computer graphics to address this knowledge gap with a computer-science mindset. Computer graphics is extremely powerful in creating content for the virtual world. The connection is therefore a natural one as the digital fabrication hardware is starving for innovative content. In this talk, I will focus on inverse design, a powerful paradigm of content synthesis for digital fabrication, which creates fabricable designs given the desired performances. Specifically, I will discuss a class of inverse design problems that deals with data-driven neural surrogate models. These surrogates learn and replace a forward process, such as a computationally heavy simulation.

Bio: Vahid Babaei leads the AI aided Design and Manufacturing group at the Computer Graphics Department of the Max Planck Institute for Informatics in Saarbrücken, Germany. He was a postdoctoral researcher at the Computational Design and Fabrication Group of Computer Science and Artificial Intelligence Laboratory (CSAIL) at MIT. He obtained his PhD in Computer Science from EPFL. Vahid Babaei is the recipient of the 2023 Germany-wide Curious Mind Award in the area of ‘AI, Digitalization, and Robotics’, the Hermann Neuhaus Prize of the Max Planck Society, and two postdoctoral fellowships awarded by the Swiss National Science Foundation. He is interested in developing original computer science methods for both engineering design and advanced manufacturing.

Talk 3 (1 pm -1:40 pm) Sven Koenig, computer science department, University of Southern California

Title: Multi-Agent Path Finding and Its Applications

Abstract: The coordination of robots and other agents becomes more and more important for industry. For example, on the order of one thousand robots already navigate autonomously in Amazon fulfillment centers to move inventory pods all the way from their storage locations to the picking stations that need the products they store (and vice versa). Optimal and even some approximately optimal path planning for these robots is NP-hard, yet one must find high-quality collision-free paths for them in real-time. Algorithms for such multi-agent path-finding problems have been studied in robotics and theoretical computer science for a longer time but are insufficient since they are either fast but of insufficient solution quality or of good solution quality but too slow. In this talk, I will discuss different variants of multi-agent path-finding problems, cool ideas for both solving them and executing the resulting plans robustly, and several of their applications. Our research on this topic has been funded by both NSF and Amazon Robotics.

Bio: Sven Koenig is a professor of computer science at the University of Southern California. Most of his current research focuses on planning for single agents (such as robots) or multi-agent systems. Additional information about him can be found on his webpages: idm-lab.org.

Talk 4 (1:40 pm -2:20 pm) Jiaoyang Li, Robotics Institute, Carnegie Mellon University

Title: Layout Design for Large-Scale Multi-Robot Coordination

Abstract: Today, thousands of robots are navigating autonomously in warehouses, transporting goods from one location to another. While numerous planning algorithms are developed to coordinate robots more efficiently and robustly, warehouse layouts remain largely unchanged – they still adhere to the traditional pattern designed for human workers rather than robots. In this talk, I will share our recent progress in exploring layout design and optimization to enhance large-scale multi-robot coordination. I will first introduce a direct layout design method, followed by a method to optimize layout generators instead of layouts. I will then extend these ideas to virtual layout design, which does not require changes to the physical world that robots navigate and thus has the potential for applications beyond automated warehouses.

Bio: Jiaoyang Li is an assistant professor at the Robotics Institute of CMU School of Computer Science. She received her Ph.D. in computer science from the University of Southern California (USC) in 2022. Her research interests lie in the coordination of large robot teams. Her research received recognition through prestigious paper awards (e.g., best student paper, best demo, and best student paper nomination at ICAPS in 2020, 2021, and 2023, along with the best paper finalist at MRS in 2023) and competition championships (e.g., winners of NeurIPS Flatland Challenge in 2020 and Flatland 3 in 2021, as well as the League of Robot Runners sponsored by Amazon Robotics in 2023). Her Ph.D. dissertation also received the best dissertation awards from ICAPS, AAMAS, and USC in 2023.

February 2, 2024

We are thrilled to announce that a special VCR event: the SFU-UBC Visual Computing Meeting will be held in SFU Burnaby campus (Feb 2nd from 10:00 am-3:00 pm). This special session will feature professors and students from the UBC and SFU visual computing community for a day of engaging discussions and networking.

Schedule – 10:00am → 12:00pm: talks (Section1 and 2) – 12:00pm → 2:00pm: lunch and posters – 2:00pm → 3:00pm: talks (Section 3)

January 26, 2024

Title: FRONTIERS IN EMBODIED AI FOR AUTONOMOUS DRIVING

Abstract: Over the last decade, fundamental advances in AI have driven unprecedented progress across many disciplines and applications. And yet, despite significant progress, autonomous vehicles are still far from mainstream even after billions of dollars of investment. In this talk we’ll explore what’s been holding progress back, and how by adopting a modern embodied AI approach to the problem, Wayve is finally unlocking the potential of autonomous driving in complex and unstructured urban environments such as central London. We’ll also explore some of our latest research in multimodal learning to combine the power of large language models with the driving problem (“LINGO-1”), and in generative world models as learned simulators trained to predict the future conditioned on ego action (“GAIA-1").

Bio: Jamie Shotton is a leader in AI research and development, with a track record of incubating transformative new technologies and experiences from early stage research to shipping product. He is Chief Scientist at Wayve, building foundation models for embodied intelligence, such as GAIA and LINGO, to enable safe and adaptable autonomous vehicles. Prior to this he was Partner Director of Science at Microsoft and head of the Mixed Reality & AI Labs where he shipped foundational features including body tracking for Kinect and the hand- and eye-tracking that enable HoloLens 2’s instinctual interaction model. He has explored applications of AI in autonomous driving, mixed reality, virtual presence, human-computer interaction, gaming, robotics, and healthcare.

He has received multiple Best Paper and Best Demo awards at top-tier academic conferences, and the Longuet-Higgins Prize test-of-time award at CVPR 2021. His work on Kinect was awarded the Royal Academy of Engineering’s gold medal MacRobert Award in 2011, and he shares Microsoft’s Outstanding Technical Achievement Award for 2012 with the Kinect engineering team. In 2014 he received the PAMI Young Researcher Award, and in 2015 the MIT Technology Review Innovator Under 35 Award. He was awarded the Royal Academy of Engineering’s Silver Medal in 2020. He was elected a Fellow of the Royal Academy of Engineering in 2021.

January 15, 2024

Congratulations to Professor Richard Zhang on being recognized as a 2024 IEEE Fellow for his “contributions to shape analysis and synthesis in visual computing.” IEEE is the world's largest technical professional organization dedicated to advancing technology for the benefit of humanity. It has more than 450,000 members in more than 190 countries and is a leading authority in many areas, including engineering, computing, and technology information. Less than one-tenth of one percent of IEEE members worldwide are selected as Fellows in any year. Fellow status is awarded to individuals with "an outstanding record of accomplishments in any of the IEEE fields of interest.”

Please check out SFU news coverage on his IEEE Fellow elevation.

November 11, 2023

SIGGRAPG Asia, the premier conference on computer graphics, will be held in Sydney, Australia this year (Dec 12-15). GrUVi lab will once again have a good show at SIGGRAPG Asia, with 5 technical papers. Congratulations to all the authors!

And here are the 5 accepted papers:

Intrinsic Harmonization for Illumination-Aware Compositing

ShaDDR: Real-Time Example-Based Geometry and Texture Generation via 3D Shape Detailization and Differentiable Rendering

CLIPXPlore: Coupled CLIP and Shape Spaces for 3D Shape Exploration

Interaction-Driven Active 3D Reconstruction with Object Interiors

Neural Packing for Real: from Visual Sensing to Reinforcement Learning

November 6, 2023

NeurIPS, the premier conference on machine learning, will be held in New Orleans this year (Dec 10-16). GrUVi lab will once again have a good show at NeurIPS, with 7 technical papers and 1 dataset and benchmarks paper!

Please refer to our publication page to see more details.

November 3, 2023

Click this link to see the talk replay.

Title: Co-speech gesture generation

Abstract: Gestures accompanying speech are essential to natural and efficient embodied human communication. The automatic generation of such co-speech gestures is a long-standing problem in computer animation. It is considered an enabling technology in film, games, virtual social spaces, and interaction with social robots. The problem is made challenging by the idiosyncratic and non-periodic nature of human co-speech gesture motion, and by the great diversity of communicative functions that gestures encompass. Gesture generation has seen surging interest recently, owing to the emergence of more and larger datasets of human gesture motion, combined with strides in deep-learning-based generative models that benefit from the growing availability of data. This talk will review co-speech gesture generation research development, focusing on deep generative models.

Bio: Dr. Taras Kucherenko is currently a Research Scientist at Electronic Arts. He finished a Ph.D. at the KTH Royal Institute of Technology in Stockholm in 2021. His research is on machine learning models for non-verbal behavior generation, such as hand gestures and facial expressions. For his research papers, he received ICMI 2020 Best Paper Award and IVA 2020 Best Paper Award. Taras was also the main organizer of The GENEA (Generation and Evaluation of Non-verbal Behavior for Embodied Agents) Workshop and Challenge in 2020, 2021, 2022, and 2023.