Matej Kristan, PhD
Associate Professor
Vice chair of Department of artificial intelligence
matej.kristan@fri.uni-lj.si
Teaching
- Machine perception(Umetno zaznavanje)
- Multimedia systems (Multimedijski sistemi)
- Advanced computer vision methods: (Napredne metode računalniškega vida)
- Machine vision (Strojni Vid): Doctoral study, FE, UL
Old courses:
- Komunikacije v Avtomatiki, (computer networks) 2009-2012: vaje, dodiplomski študij, FE, UL
- Sistemi Daljinjskega vodenje (wireless sensor networks) 2009-2012: vaje, dodiplomski študij, FE, UL
- Strojni Vid (machine vision) 2010-2012: podiplomski študij, FE, UL
- Večpredstavitveni sistemi (computer vision for multimedia sistems): vaje z letom 2010/2011, dodiplomski študij, FRI, UL
- Računalniško zaznavanje (computer vision): vaje z letom 2010/2011, dodiplomski študij, FRI, UL
- Multimedijski sistemi (computer vision for multimedia systems): vaje z letom 2011/2012, dodiplomski študij, FRI, UL
- Umetno zaznavanje (computer vision): vaje z letom 2011/2012, dodiplomski študij, FRI, UL
- Robotics and machine perception 2012/2013: predavanja(Robotika in računalniško zaznavanje)
Events & Service to community
- Associate editor at International Journal of Computer Vision, IJCV 2021-
- Area chair at IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) 2022
- Organization committee of the ninth Visual Object Tracking Challenge Workshop VOT2021 in conjunction with the ICCV2021
- Organization committee of the eighth Visual Object Tracking Challenge Workshop VOT2020 in conjunction with the ECCV2020
- Organization committee of the seventh Visual Object Tracking Challenge Workshop VOT2019 in conjunction with the ICCV2019
- Organization committee of the sixth Visual Object Tracking Challenge Workshop VOT2018 in conjunction with the ECCV2018
- Organization committee of the fifth Visual Object Tracking Challenge Workshop VOT2017 in conjunction with the ICCV2017
- Program chair of the 10th Int'l Symposium on Image and Signal Processing and Analysis, ISPA2017 (link to page)
- Organization committee of the fourth Visual Object Tracking Challenge Workshop VOT2016 in conjunction with the ECCV2016 (link to page)
- Organization committee of the third Visual Object Tracking Challenge Workshop VOT2015 in conjunction with the ICCV2015 (link to page)
- Competition chair at the 11th IEEE International Conference on Automatic Face and Gesture Recognition, FG2015 (link to page)
- Organization committee of the second Visual Object Tracking Challenge Workshop VOT2014 in conjunction with the ECCV2014 (link to page)
- Organization committee of the first Visual Object Tracking Challenge Workshop VOT2013 in conjunction with the ICCV2013 (link to page)
- Organization committee of the Pattern Recognition section at the 22nd International Electrotechnical and Computer Science Conference ERK2013
- Organization committee of the 17th Computer Vision Winter Workshop CVWW2012 (link to page)
Research visits
- 2020 (summer) Visiting professor at Center for Machine Perception, Czech Technical University, Prague
- 2018 (summer) Visiting professor at Laboratory of Signal Processing, Tampere University of Technology, Finland
- 2016 (winter) Visiting researcher at School of Computer Science, University of Birmingham, UK
- 2013 (winter) Visiting researcher at School of Computer Science, University of Birmingham, UK
Awards
- 2020 Our work on RGB/RGBD long-term tracking and evaluation was selected as one of the Excellent Slovenian research achievements in 2020 by the Slovenian Research Agency.
- 2020 Golden plaque award for outstanding scientific achievements of a research group awarded by University of Ljubljana (https://www.uni-lj.si/tedenuniverze/zlata_plaketa/).
- 2020 Two PhD students, Borja Bovcon (under by PhD supervision), and Vitjan Zavrtanik (co-supersvised his research) received an award for outstanding research achievements of PhD students in 2020 (three awards given), awarded by the Faculty of computer and information science, UNI-LJ (link to page).
- 2020 Two PhD students Domen Tabernik (PhD under my co-supervision) and Jon Muhovič (co-supervised his research) received an honorable mention of the research achievements of PhD students in 2020, awarded by the Faculty of computer and information science, UNI-LJ (five awards given).
- 2020 Runner-up for supervisor of the year award organized by Mlada Akademija (among all nominations from Slovenia Universities and Institutes).
- 2019 Our work on deformable object tracking was selected as one of the Excellent Slovenian research achievements in 2019 by the Slovenian Research Agency (link to page).
- 2019 two of my PhD students (Alan Lukežič and Borja Bovcon) received an award for outstanding research achievements of PhD students in 2019 (four awards given), awarded by the Faculty of computer and information science, UNI-LJ (link to page).
- 2018 Golden plaque award for outstanding scientific and pedagogic contributions awarded by University of Ljubljana (link to page)
- 2018 Award for excellent pedagogical work, awarded by the Faculty of computer and information science, University of Ljubljana
- 2018 two of my PhD students (Alan Lukežič and Domen Tabernik) received an award for outstanding research achievements of PhD students in 2018 (five awards given), awarded by the Faculty of computer and information science, UNI-LJ (link to page).
- 2018 my PhD student Borja Bovcon received an honorable mention of the research achievements of PhD students in 2018, awarded by the Faculty of computer and information science, UNI-LJ (link to page).
- 2018 my student Lojze Žust received Prešeren award from the Faculty of computer and information science, UNI-LJ, for his bachelor thesis (link to page).
- 2017 Coauthor of the work on VOT initiative that was selected as one of 10 exceptional research achievements in 2017 awarded by University of Ljubljana.
- 2017 my student Borja Bovcon received Prešeren award from the Faculty of computer and information science, UNI-LJ, for his masters thesis "Improved segmentation model for robotic boat obstacle detection".
- 2015 my student Alan Lukežič received Prešeren award from the Faculty of computer and information science, UNI-LJ, for his masters thesis "Improved robust part-based model for visual object tracking".
- 2015 Best paper award at pattern recognition section at ERK2015 for the paper "Filtering out nondiscriminative keypoints by geometry-based keypoint constellations" authored by my student Domen Rački.
- 2013 Young University teachers and researchers award for excellent teaching and research achievements, awarded by University of Ljubljana.
- 2013 coauthor of the "Robust visual tracking using an adaptive coupled visual model" (link to page) for which the exceptional scientific achievements in 2012 award was awarded by the Slovenian research agency.
- 2012 Contributed on the research project "A system for interactive learning in dialogue with the tutor", for which an award for exceptional research achievements in 2011 was attributed by the Slovenian research agency to the research program P2-0214.
- 2010 Award for contributions in popularization of Computer and information science, awarded by the Faculty of Computer and Information Science, University of Ljubljana
- 2009 Vodovnikova nagrada - Vodovnik Award for excellent research achievements relating to doctoral thesis, Faculty of Electrical Engineering, University of Ljubljana (link to page)
- 2009 Award for excellent research achievements in the last years, Faculty of Computer and Information Science, University of Ljubljana
- 2009 Coauthor of “Intelligent security system for surveillance of small areas”, for which Innovation award for applicative project was awarded by the Innovation forum, sponsored by Slovenian business and foreign investment agency.
- 2005 Best paper award at the International Symposium on Image and Signal Processing and Analysis ISPA2005, for the paper “Multiple interacting targets tracking with application to team sports”.
Research
My primary research interests lie in computer vision and machine learning. I am particularly interested in visual tracking, object detection and recognition. In the field of machine learning and pattern recognition, I have published several works on univariate/multivariate batch/online kernel density estimation. My secondary fields of interest are computer vision for mobile robotics and computer-vision-enabled multimedia systems.
Measures of Camera Focus
This page deals with the problem of how to measure the extent of camera focus direclty from the (part of) observed image. We propose frequency-based measure, which uses DCT to extract the the image spectrum and uses a Bayes entropy of the spectrum to measure the focus.
Visual Tracking
Discriminative Correlation Filters are very popular in visual object tracking due to the efficient implementation and great tracking performance results. Here we present several improvements in discriminative correlation filters.
One of the problems of visual tracking evaluation is a lack of a consistent evaluation methodology. This is hampering the cross-paper tracker comparison and faster advancement of the field. In our research we investigate different aspects of tracking evaluation. A continuous effort that is a part of our work is also the Visual Object Tracking Challenge (VOT).
This page deals with the problem of how to track objects as they undergo non-rigid deformations, and as parts become invisible. The page contains some preliminary work on a combined local and global visual model, that simultaneously changes its structure by adapting to the potentially nonrigid target and localizing it in the image.
This page deals with the problem of how to build and maintain a color-based visual model for tracking in cluttered environment like sports court. We propose a histogram based visual model which incorporates the background and propose a measure of presence for probabilistic tracking.
This page deals with the problem of how to track an object while under (partial) occlusion by another visually similar object. We show how the Lucas-Kanade optical flow can be used to derive a local-motion visual model, how to efficiently update the model during tracking and propose a probabilistic model that can be used within a particle filter.
This page deals with the problem of how to efficiently model target’s dynamics for visual tracking. We propose a dynamic model that allows tracking with a very small number of particles (25 in experiements) in the particle filter while improving tracking accuracy and rebustness over related dynamic models.
This page deals with the problem of how to track multiple visually similar targets in sports. The page is a collection of solutions to tackle this problem with a particle filter. One solution is using the local-motion-based tracking, the other is taylored for constrained environments and uses Voronoi diagrams to reduce the problem of estimating a high-dimensional state with the particle filter.
Incremental Mixture Models
This page deals with the problem of how to incrementally build a generative model from data using a positive as well as negative examples. We propose an incremental update rule for building one dimensional Gaussian mixture models based on Kernel Density Estimation that support learning from negative examples.
This page deals with the problem of how to design an algorithm that would use as little assumptions as possible about the input data and would allow building a generative model by observing only a single or a few data points at a time. The algorithm is based on multivariate Kernel Density Estimator which uses a revitalization scheme and is robust to data ordering.
We present an algorithm for building a discriminative models by observing only a single or a few data points at a time. The algorithm is and extension of the multivariate online Kernel Density Estimator, and uses a new measure of discrimination loss to determine how much a classifier can be compressed without modyfing its performance.
Deep structured nets and Hierarchical compositional models
We propose a novel deep network architecture that combines the benefits of discriminative deep learning and the benefits of compositional hierarchies. As one of the benefits we emphasize the ability to automatically adjust receptive fields to either small or large receptive fields depending on the for problem at hand and the ability to visualize deep features through explicit compositional structure.
We propose a novel approach that preserves the appealing properties of the generative hierarchical compositional models, while at the same time improves their discrimination properties. We achieve this by introducing a network of discriminative nodes on top of the existing generative hierarchy. We show that in range of ~10% of generative parts are crucial for discrimination.
As extension to LHOP model we have developed a shape descriptor capable of using compositional parts learnt using LHOP model to provide a descriptor that is compatible with HOG descriptor and can be easily used as direct replacement.
We deal with a problem of Multi-class Object Representation and present a framework for learning a hierarchical shape vocabulary capable of representing objects in hierarchical manner using a statistically important compositional shapes. The approach takes simple oriented contour fragments and learns their frequent spatial configurations. These are recursively combined into increasingly more complex and class specific shape compositions, each exerting a high degree of shape variability
Mobile robotics
We have designed approaches for 2D laser-range-data-based room categorization that are grounded on a compositional hierarchical representation of space. We have also developed a part-based image representation that is suitable for robust vision-based room categorization.
Unmnanned surface vehicles (USV) are robotic boats that can be used for coastal patrolling in a numerous applications ranging from surveillance to water cleanness control. We are developing computer vision algorithms that enable autonomous operation in the highly dynamic environments in which the USVs are applied.
Downloads
 |
This is a Matlab demo code for the semantic segmentation model for obstacle image map estimation for unmanned surface vehicles. The demo requires downloading the MOD dataset and has pretrained hiperparameters on the MOD dataset.
|
|
NEW! Version 3.5 This is a Matlab research code that is based on the papers on Online Kernel Density Estimation with Gaussian Kernels and Online Discriminative Kernel Density Estimation with Gaussin Kernles. The code essentially demonstrates estimation of a Gaussian Mixture Model from a stream of data.
|
 |
The code is a minimal implementation of a batch kernel density estimator. Since the code is based on our new bandwidth estimator, it allows KDE construction even from preclustered/compressed sets of samples and weighted data. This is also a minimal demonstration of the general bandwidth estimator proposed in “Online Kernel Density Estimation with Gaussian Kernels”.
|
 |
This is a research code for 2D KDE that is based on the paper “Online Kernel Density Estimation with Gaussian Kernels”.
|
 |
This is a demo code for the unscented Hellinger distance between a pair of Gaussian mixture models. The code follows the derivation of the multivariate unscented Hellinger distance introduced in [1]. Unlike the Kullback-Leibler divergence, the Hellinger distance is a proper metric between the distributions and is constrained to interval (0,1) with 0 meaning complete similarity and 1 complete dissimilarity.
|
 |
This is a Matlab implementation of the Bayes spectral based measure of camera focus using a discrete cosine transform.
|
Projects
Please refer to our project page for full list of project I am/was cooperating in. In the following only projects PI-ed by me are listed:
The project primary goal is to develop the next-generation maritime environment perception methods, which will harvest the power of end-to-end trainable deep models for essential challenges of safe operation like: general obstacle detection with re-identification, implicit detection of hazardous areas and sensor fusion for improved detection.
The project primary goal is to develop functionalities required for robust autonomous navigation of USVs in uncontrolled environments, primarily relying on the captured visual information. The project focuses on obstacle detection using monocular and stereo systems, development of efficient visual tracking algorithms for marine environments and environment representation through sensor fusion.
This project primary goal is to develop advanced machine learning methods for rapid sea level forecasting. This is a collaboration with climatologists and geophysicists from the Slovenian environmental agency (ARSO). The first result of the project was a deep architecture Hidra 1.0 which considers historical/predicted weather parameters and historical sea levels and delivers sea level predictions on par (and in some cases better) with an operational numerical model at a fraction of computational cost (~half million times faster). Hidra v.1.0 has been made live and operational by ARSO since May 2021.
The main goal of this project is development of computer-vision-based automated counters applicable to the domain of scyphistoma census in underwater
imagery. Such counters are crucial for processing extremely large datasets, vastly reducing the required manual labor and facilitating census orders of magnitude grater than what is possible with today's semi-manual techniques. The methods apply learning-based methodology, allowing to train a general polyp counter applicable to a large variety of images as well as training for a specific type of images to maximize a task-specific detection performance.
We have developed algorithms for robust and fast dent detection and characterization on reflective surfaces that does not require reference fringe-pattern images.
Visual tracking algorithms used for tracking with drones.
Journal publications
Book chapters
Selected conference publications
All publications
Here's a local list of all my relevant publications (journals+conferences)
A most up-to-date list of my publications (journals+conferences+theses+etc.) is avalible from the SICRIS
PhD Students
Here's a list of PhD students under my supervision:
- Alan Lukežič, Area of research: Visual object tracking, (active)
- Domen Tabernik, Area of research: Deep compositional models, (active)
- Borja Bovcon, Area of research: Vision for unmanned surface vehicles, (active)
- Nejc Dougan, Area of research: 3D large area segmentation, (active)
- Peter Uršič, A Compositional Hierarchical Architecture for Spatial Modelling, (finished 2016)
- Luka Čehovin, A hierarchical adaptive model for robust short-term visual tracking, (finished 2015)
Some of the students (Bsc/Msc) have published video presentations of their theses at the ViCoS teaching page.