Greg MoriProfessor and Director
School of Computing Science
Simon Fraser University
8888 University Drive
CANADA V5A 1S6
Office: TASC1 8007
Phone: (778) 782-7111
Fax: (778) 782-3045
Ph.D. in Computer Science, University of California at Berkeley, 2004.
Hon. B.Sc. in Computer Science and Mathematics, University of Toronto, 1999.
Dr. Greg Mori was born in Vancouver and grew up in Richmond, BC. He
received the Ph.D. degree in Computer Science from the University of California,
Berkeley in 2004. He received an Hon. B.Sc. in Computer Science
and Mathematics with High Distinction from the University of Toronto in 1999.
He spent one year (1997-1998) as an intern at Advanced Telecommunications Research
(ATR) in Kyoto, Japan. After graduating from Berkeley, he
returned home to Vancouver and is currently a Professor in
the School of Computing Science at Simon Fraser University.
He was a Visiting Scientist at Google in Mountain View, California in 2014-2015. Returning to Vancouver again, he became Director of the School of Computing Science in May 2015.
Dr. Mori conducts research in computer vision and machine learning, and teaches classes in data structures and programming, artificial intelligence, computer vision, and machine learning. He serves on the editorial boards of IJCV and T-PAMI, the top journals in computer vision, and on the organizing committees for CVPR, ICCV, and ECCV, the top conferences in computer vision. He is privileged to have worked with many excellent students while at SFU.
My research is in computer vision, and is concerned with developing
algorithms that automatically interpret images and videos,
particularly those containing people. I have made significant
contributions towards solving the problems of human pose estimation
and human action recognition. At a broad level, the methodology
followed is to construct features and representations that capture our
intuition regarding these vision problems. We operationalize these via
machine learning algorithms, adapting them to suit our
Specific examples of features and representations include work on superpixels for representing images, motion features for human action recognition, and bag-of-words models for video sequences. We have developed variants of machine learning algorithms for models such as latent SVMs, hidden Conditional Random Fields (hCRFs), and Latent Dirichlet Allocation (LDA) to implement these ideas.
Research interests keywords:
|Hossein Hajimirsadeghi (PhD) had a paper accepted to IEEE International Conference on Computer Vision (ICCV), 2015. The paper presents a boosting method for combining multiple high-capacity models in a structured prediction setting, demonstrating results on a variety of applications including group activity recognition and 3d action recognition.|
|Zhiwei Deng (MSc), Mengyao Zhai (MSc), Lei Chen (MSc), Yuhao Liu (BSc), and Srikanth Muralidharan (MSc) had a paper accepted to the British Machine Vision Conference (BMVC), 2015. A deep learning framework for group activity recognition via message passing is presented.|
|Wang Yan (PDF) and Jordan Yap (BSc) had a paper accepted to the British Machine Vision Conference (BMVC), 2015. Video retrieval in a "one-shot" (single example) setting is improved by leveraging multi-task learning from existing models.|
|Mehran Khodabandeh (PhD), Hossein Hajimirsadeghi (PhD), Arash Vahdat (URA), and Guang-Tong Zhou (PhD) had a paper accepted to Workshop on Group and Crowd Behavior Analysis and Understanding at CVPR. Human interaction discovery in surveillance video is addressed using a human-in-the-loop clustering framework.|
|Hossein Hajimirsadeghi (PhD), Arash Vahdat (URA), and Wang Yan (PDF) had a paper accepted to IEEE Computer Vision and Pattern Recognition (CVPR), 2015. The paper presents a method for multiple instance learning with cardinality potential kernels, with applications to group activity recognition, video summarization, and internet video event detection.|
|Yasaman Sefidgar (MSc) and Arash Vahdat (URA) had a paper accepted to Computer Vision and Image Understanding (CVIU). Human-human and human-vehicle interactions are detected in surveillance video using a discriminative model.|
|Mengyao Zhai (MSc), Lei Chen (MSc), Mehran Khodabandeh (PhD), and Jinling Li, had a paper accepted to IAPR Machine Vision Applications (MVA), 2015. Dense trajectory features are used in a regression model for vehicle and person detection in videos.|
|Mehran Khodabandeh, successfully defended his M.Sc. thesis Discovering Human Interactions in Videos with Limited Data Labeling, Congratulations Mehran!|
|Arash Vahdat, successfully defended his Ph.D. thesis Weakly Supervised Models For Recognizing And Clustering High-Level Complex Events In Video. Congratulations Arash!|
|Nataliya Shapovalova successfully defended her Ph.D. thesis Towards Action Recognition and Localization in Videos with Weakly Supervised Learning. Congratulations Nataliya!|
|Jinling Li successfully defended her M.Sc. thesis Road User Detection and Analysis in Traffic Surveillance Videos. Congratulations Jinling!|
|Arash Vahdat (PhD) and Guang-Tong Zhou (PhD) had a paper accepted to the European Conference on Computer Vision (ECCV), 2014. An approach for discovering clusters of related internet videos is presented, utilizing noisy tag labels and image features.|
|Yasaman Sefidgar successfully defended her M.Sc. thesis Discriminative Key-Segment Model for Interaction Detection. Congratulations Yasaman!|
|Amir Bakhtiari successfully defended his M.Sc. thesis Detecting Pedestrians Using Motion Patterns: A Latent Tracking Approach. Congratulations Amir!|
|Nataliya Shapovalova (PhD) had a paper accepted to Neural Information Processing Systems (NIPS), 2013. A joint model for visual saliency and action recognition in videos is described, trained in a latent SVM framework utilizing eye gaze data.|
|Guang-Tong Zhou (PhD), Tian Lan (PhD), and Arash Vahdat (PhD) had a paper accepted to Neural Information Processing Systems (NIPS), 2013. The paper describes an approach for clustering with latent variables under a max-margin criterion, with application to videos of human activities.|
|Tian Lan successfully defended his Ph.D. thesis From Flat to Hierarchical: Modeling Structures in Visual Recognition. Congratulations Tian!|
|Arash Vahdat (PhD) had a paper accepted to IEEE International Conference on Computer Vision (ICCV), 2013. An algorithm for predicting a set of tags that describe and image or video is described, trained from noisy tag annotations.|
|Arash Vahdat (PhD) and Kevin Cannons (PDF) had a paper accepted to IEEE International Conference on Computer Vision (ICCV), 2013. A compositional model for video event recognition is presented, based on a novel multiple kernel learning algorithm that incorporates latent variables.|
|Tian Lan (PhD) had a paper accepted to IEEE International Conference on Computer Vision (ICCV), 2013. A multi-level object detection framework that models sub-categories through to visual composites is developed.|
|Hossein Hajimirsadeghi (PhD) and Jinling Li (MSc) had a paper accepted to the Conference on Uncertainty in Artificial Intelligence (UAI), 2013. A multiple instance learning algorithm is developed, which models varying levels of positive instances in a bag. A discriminative training algorithm is proposed, based on efficient inference of cardinality-based clique potential functions.|
|Tian Lan (PhD) had a paper accepted to IEEE Computer Vision and Pattern Recognition (CVPR), 2013. This work develops an image tag ranking algorithm that can determine which keywords are more relevant to a given image.|
|Guang-Tong Zhou (PhD), Tian Lan (PhD), and Weilong Yang had a paper accepted to IEEE Computer Vision and Pattern Recognition (CVPR), 2013. An image classification algorithm that matches a set of objects to an image is developed.|
|Yuke Zhu (BSc) and Tian Lan (PhD) had a paper on activity recognition accepted to IAPR Conference on Machine Vision Applications (MVA), 2013. Latent variable models were used to analyze nursing home surveillance video.|
Are you a prospective student? Please read this page for prospective students.