This tutorial aims at promoting discussions among researchers investigating innovative second-order (bilinear), kernel and tensor-based approaches to computer vision problems. Specifically, we will stimulate discussions on recent advances, ongoing developments, and novel applications of bilinear, kernel and multilinear algebra, optimization, and feature representations using matrices and tensors in the context of CNN learning.
TOPICS
We have addressed a wide range of theoretical and practical issues including, but not limited to the following topics:
- Bilinear, kernel and tensor methods in low-level feature design and deep learning
- Power Normalisation, non-linearities and their formulations
- Mid-level representations with co-occurrence matrices and tensors
- Low-rank factorisation methods and denoising approaches
- Latent topic models using matrices, kernels and tensor methods
- Co-occurrence matrices, kernels and tensors in optimization and dictionary learning
- Advancements in Riemannian geometry, kernel methods and multilinear algebra
- Dimensionality reduction, similarity learning, metric learning, and other machine learning topics
- Applications of co-occurrences, kernels and tensors for:
- Object recognition
- Scene understanding
- Fine-grained classification
- Action recognition
- Industrial and medical applications
- Other CV and ML problems
- Other related topics not listed above
SCHEDULE
Below is the program of the tutorial that takes place on the 4th of December, 2022. Below is the Detailed Program with abstracts and biographies of our speakers (or click on links in tables).
Afternoon Session
Time | Speaker | Title |
---|---|---|
14:00 | Organizers | Welcome |
14:11 | Dr. Naila Murray | Invited Talk I: Normalization and reweighting techniques for effectively aggregating higher-order local visual representations. |
15:00 | A/Prof. Mehrtash Harandi | Invited Talk II: Poincaré Kernels for Hyperbolic Representations. |
15:51 | Short break | Venue |
16:00 | Prof. Ruiping Wang | Invited Talk III: Riemannian Metric Learning and its Vision Applications. |
16:51 | Dr. Piotr Koniusz | Invited Talk IV: Understanding High Order Pooling. |
17:40 | Organizers | Closing remarks |
INFORMATION
- Kindly note that registration at ACCV'22 webpage is mandatory for everyone participating in the workshop (at least a registration for the tutorial).
- The workshop will take place on the 4th of December, 2022.
- You can find an archive listing related papers on second-order pooling here.
DETAILED PROGRAM
Below is the list of speakers who will give a talk during the tutorial (including organizers):- Dr. Naila Murray (Facebook AI Research)
Title: Normalization and reweighting techniques for effectively aggregating higher-order local visual representations.Abstract: Vectorial representations for images, videos, and other visual content, are usually derived from local representations that correspond to sub-regions of the input signal. In this talk, I will discuss recent work on generating higher-order local representations that are more discriminative when compared with lower-order representations. Because these representations are often high-dimensional and thus resource intensive (both in compute and memory) I will also discuss aggregation methods that, in combination with suitable normalisation and reweighing techniques, obtain more efficient representations that retain discriminative power. Lastly, I will present quantitative evaluations that demonstrate the effectiveness of the resulting representations for a variety of problems, including instance-level image retrieval and object detection.
Biography: Naila Murray obtained a BSE in electrical engineering from Princeton University in 2007. In 2012, she received her Ph.D. from the Universitat Autonoma de Barcelona, in affiliation with the Computer Vision Center. She joined Xerox Research Centre Europe in 2013 as a research scientist in the computer vision team, working on topics including fine-grained visual categorization, image retrieval and visual attention. From 2015 to 2019 she led the computer vision team at Xerox Research Centre Europe, and continued to serve in this role after its acquisition and transition to becoming NAVER LABS Europe. In 2019, she became the director of science at NAVER LABS Europe. In 2020, she joined Facebook AI Research where she is currently a senior research engineering manager. She has served as area chair for ICLR 2018, ICCV 2019, ICLR 2019, CVPR 2020, ECCV 2020, and CVPR 2022, and program chair for ICLR 2021. Her current research interests include few-shot learning and domain adaptation. - Prof. Ruiping Wang (Chinese Academy of Sciences)
Title: Riemannian Metric Learning and its Vision Applications.Abstract: The talk mainly focuses on modeling videos/image sets using second-order SPD matrices and learning discriminative metrics on the SPD manifold for visual recognition, with applications to image set classification, video-based face recognition/retrieval, dynamic emotion recognition, action recognition.
Biography: Ruiping Wang is a Professor at the Institute of Computing Technology (ICT), Chinese Academy of Sciences (CAS). Prior to joining ICT in July 2012, He worked as a post-doctoral researcher with Tsinghua University from 2010 to 2012, and a research associate with the University of Maryland at College Park from 2010 to 2011. He has published more than 100 papers in peer-reviewed journals and conferences, including IEEE TPAMI, TIP, IJCV, CVPR, ICCV, ECCV, ICML. Dr. Wang serves as an Associate Editor for Neurocomputing (Elsevier), Pattern Recognition, Area Chair for IEEE WACV'18/19/20/22/23, ICME19/20/23, CVPR21/22, ICCV21, ECCV22 and ACCV22. He has co-organized tutorials in ACCV14/CVPR15/ECCV16/ICIP17/ICCV19. His current research focuses on projects related to visual scene understanding and human face analysis. He is a senior member of the IEEE. - Dr. Mehrtash Harandi (Monash University)
Title: Poincaré Kernels for Hyperbolic Representations.Abstract: Embedding data in hyperbolic spaces has proven beneficial for many advanced machine learning applications such as image classification and word embeddings. However, working in hyperbolic spaces is not without difficulties as a result of its curved geometry. In Euclidean spaces, one can resort to kernel machines that not only enjoy rich theoretical properties but also can lead to a superior representational power (eg, infinite-width neural networks). In this talk, we introduce positive definite kernel functions for hyperbolic spaces. This brings in two major advantages, 1. kernelization will pave the way to seamlessly benefit from kernel machines in conjunction with hyperbolic embeddings, and 2. The rich structure of the Hilbert spaces associated with kernel machines enables us to simplify various operations involving hyperbolic data. That said, identifying valid kernel functions is not straightforward and is indeed considered an open-problem in the learning community. Our work here addresses this gap and develops several valid positive definite Poincaré kernels for hyperbolic representations, including the universal ones (eg, RBF). We study the effectiveness of the proposed hyperbolic kernels on a variety of challenging tasks including few-shot learning and zero-shot learning.
Biography: A/Prof. Mehrtash Harandi is with the Department of Electrical and Computer Systems Engineering at Monash University. He is also a contributing research scientist in the Machine Learning Research Group (MLRG) at Data61/CSIRO and an associated investigator at the Australian Center for Robotic Vision (ACRV). His current research interests include theoretical and computational methods in machine learning, computer vision, signal processing, and Riemannian geometry. - Dr. Piotr Koniusz (Data61/CSIRO and Australian National University)
Title: Understanding High Order Pooling.Abstract: In this talk, I will focus on understanding the role of pooling, especially in the context of estimating second-order matrices and tensors. I will discuss the notion of bursiness, typical first- and high-order pipelines, relation of the so-called Power Normalisation to non-Euclidean distances. I will motivate the need for robust estimation of statistics. Subsequently, I will talk about our work on taxonomy of Power Normalizations and their relation to the reverse heat diffusion. Finally, I will briefly illustrate several pipelines that build on high order pooling.
Biography: Dr. Koniusz is a senior research scientist in Machine Learning Research Group at Data61/CSIRO (former NICTA). He is also a senior honorary lecturer at Australian National University (ANU). Previously, he worked as a post-doctoral researcher in the team LEAR, INRIA, France. He received my BSc degree in Telecommunications and Software Engineering in 2004 from the Warsaw University of Technology, Poland, and completed his PhD degree in Computer Vision in 2013 at CVSSP, University of Surrey, UK. His interests include statistical representation learning. - Dr. Peyman Moghadam (Data61/CSIRO)
Biography: Dr. Peyman Moghadam is a Principal Research Scientist at CSIRO Data61, Adjunct Professor at the Queensland University of Technology (QUT), and Adjunct Associate Professor at the University of Queensland (UQ). He is leading the Embodied AI research cluster at the CSIRO Robotics and Autonomous Systems group working at Intersection of Robotics and Machine learning. As the leader of Spatiotemporal portfolio at CSIRO's Machine Learning and Artificial Intelligence (MLAI) Future Science Platform, Dr Moghadam also oversees research and development of MLAI methods for scientific discovery in spatiotemporal data streams. Before joining CSIRO, Peyman worked in a number of world leading organisations such as the Deutsche Telekom Laboratories (Germany) and the Singapore-MIT Alliance for Research and Technology (Singapore). Dr Moghadam has led several large-scale multidisciplinary projects and won numerous awards for his innovations, including CSIRO Julius Career award, National, and Queensland iAward for Research and Development, and the Lord Mayor’s Budding Entrepreneurs Award. In 2019, he held a Visiting Scientist appointment at the Agricultural Robotics and Engineering group at the University of Bonn, as part of the CSIRO Julius Career Award. His current research interests include self-supervised learning for robotics, embodied AI, 3D multi-modal perception (3D++), robotics, computer vision, deep learning, and 3D thermal/hyperspectral imaging.
- Saimunur Rahman (Data61/CSIRO)
Biography: Saimunur Rahman recently completed his PhD from Data61/CSIRO and the University of Wollongong. He is currently a post-doctoral research fellow at Data61/CSIRO. His doctoral research was about higher-order visual representation learning with deep networks. He received his M.Sc. (by Research) in Computer Vision and B.Sc. in Computer Science & Engineering. He has been an awardee on many scholarships and honours throughout his life for academic and research excellence, including the Data61 PhD scholarship. His current research interests include computer vision, robotics and machine learning. He regularly reviews papers from top artificial intelligence conferences such as CVPR, ECCV, ACM MM etc., and serves on the program committees of various international conferences.
- Shan Zhang (Australian National University)
Biography: Shan Zhang is currently pursuing a PhD at Australian National University and working under the supervision of Dr. Piotr Koniusz in College of Engineering and Computer Science. She is interested in feature learning, object detection, domain adaptation and few-shot learning. Her main focus is on kernel learning features and higher-order pooling combined with few-shot learning. She has published papers in CVPR, IJCAI, ACCV. In 2017-2020, she received the M.S. degree in Software Engineering at Beijing Union University, Beijing, China. In 2013-2017, she completed the B.S. degree from Beijing Union University, majoring in Electrical Engineering and Automation.
CITATION
If you wish to cite any topics raised during the tutorial, refer to specific papers of our speakers. Additionally, you are welcome to cite the tutorial itself:@misc{highercv_tutorial_2022, title = {Higher-order Visual Representation Learning}, author = {S. Rahman and S. Zhang and P. Moghadam and P. Koniusz}, howpublished = {ACCV Tutorial, \url{https://www.koniusz.com/highercv-accv22}}, note = {Accessed: 04-12-2022}, year = {2022}, }
ORGANISERS
- Saimunur Rahman (Data61/CSIRO)
- Shan Zhang (the Australian National University)
- Dr. Peyman Moghadam (Data61/CSIRO)
- Dr. Piotr Koniusz (Data61/CSIRO and the Australian National University)