Navigation

Martin Godec

M. Godec Martin Godec

E-Mail godec(at)icg.tugraz.at
Phone +43 316 873-5031
Office Room 3.02
TUGraz Business Card Martin Godec

Research Interests Visual Tracking, (On-line) Machine Learning, Representation
Project MobiTrick

Martin Godec

View Martin Godec's profile on LinkedIn

Short CV

Martin Godec received his MSc Degree in Telematics from the Graz University of Technology in 2008. He is currently a research assistant at the Institute for Computer Graphics and Vision where he is involved in the Mobi-Trick Project. His research is focused on Object Tracking, considering On-line Learning of Representations.

Martin Godec has been Co-Chair of the Computer Vision Winter Workshop 2011.

Download full CV

Software

LibBOB-1.0: You can download our multi-class learning framework here.
HoughTrack: Tracking code for "Hough-based Tracking of Non-rigid Objects" can be found here

Research

Hough-based On-line Tracking of Non-Rigid Objects (Godec, Roth, Bischof)

Online learning for tracking has demonstrated to be a successful approach for tracking of previously unknown objects. The major limitation of such approaches is that they are limited to a bounding-box representation, which has several drawbacks like less accurate foreground background separation, handling of highly non-rigid and articulated objects etc. . In this paper, we present a novel tracking-by-detection approach based on the Generalized Hough-Transform. We extend the idea of Hough-Forests to the online domain and couple the center vote based detection and back-projection with a rough segmentation based on graph-cuts. This significantly reduces the amount of noisy training samples during online learning and effectively prevents the tracker from drifting. We demonstrate that our method successfully tracks various objects that are unknown in advance, even under heavy non-rigid transformations, partial occlusions, scale changes and rotations. In the experiments, we demonstrate our tracker delivering robust and accurate tracking results on various sequences compared to state-of-the-art methods (both bounding-box based as well as part-based).

This work has been supported by the Austrian FFG project MobiTrick (8258408) under the FIT-IT program.

Hough-based Tracking of Non-rigid Objects (Paper (pdf))

Project page

Improving Classifiers with Unlabeled Weakly-Related Videos (Leistner, Godec, Schulter, Saffari, Werlberger, Bischof)

Current state-of-the-art object classification systems are trained using large amounts of hand-labeled images. In this paper, we present an approach that shows how to use unlabeled video sequences, comprising weakly-related object categories towards the target class, to learn better classifiers for tracking and detection. The underlying idea is to exploit the space-time consistency of moving objects to learn classifiers that are robust to local transformations. In particular, we use dense optical flow to find moving objects in videos in order to train part-based random forests that are insensitive to natural transformations. Our method, which is called Video Forests, can be used in two settings: first, labeled training data can be regularized to force the trained classifier to generalize better towards small local transformations. Second, as part of a tracking-by-detection approach, it can be used to train a general codebook solely on pair-wise data that can then be applied to tracking of instances of a priori unknown object categories. In the experimental part, we show on benchmark datasets for both tracking and detection that incorporating unlabeled videos into the learning of visual classifiers leads to improved results.

This work has been supported by the Austrian FFG projects MobiTrick (8258408) and OUTLIER (820923) under the FIT-IT program as well as the HDVIP Bridgeproject (827544).

Improving Classifiers with Unlabeled Weakly-Related Videos (Paper (pdf))
Additional Material ( Video Proceedings (avi), Poster (pdf), Supplemental Material (avi) )

On-line Multi-View Forests for Tracking (Leistner, Godec, Saffari, Bischof)

A successful approach to tracking is to on-line learn discriminative classifiers for the target objects. Although these tracking-by-detection approaches are usually fast and accurate they easily drift in case of putative and self-enforced wrong updates. Recent work has shown that classifier-based trackers can be significantly stabilized by applying semi-supervised learning methods instead of supervised ones. In this paper, we propose a novel on-line multi-view learning algorithm based on random forests. The main idea of our approach is to incorporate multi-view learning inside random forests and update each tree with individual label estimates for the unlabeled data. Our method is fast, easy to implement, benefits from parallel computing architectures and inherently allows to exploit multiple views for learning from unlabeled data. In the tracking experiments, we outperform the state-of-the-art methods based on boosting and random forests.

This work has been supported by the Austrian FFG project MobiTrick (825840) and Outlier (820923) under the FIT-IT program.

On-line MV-Forests (Paper (pdf))
Additional Material ( Supplemental Material (avi), Presentation (pdf))

Audio-Visual Co-Training for Vehicle Classification (Godec, Leistner, Bischof, Starzacher, Rinner)

Traffic monitoring is a very important task within the field of surveillance and has a variety of applications such as electronic toll collection, traffic jam prediction or truck ban inspection. At the beginning of a typical monitoring chain stands the successful detection and classification of vehicles which is then used for further high-level reasoning. Static classification systems that are applied in practice usually demand human calibration and are based on large amounts of hand-labeled data in order to allow for 24-hour operation under harsh environmental conditions. This makes these static systems costly and handicaps flexibility in terms of maintenance and deployment. In this paper, we introduce a fully autonomous vehicle classification system that continuously learns from large amounts of unlabeled data. For that purpose, we propose a novel on-line co-training method based on visual and acoustic information. Our system does not need complicated microphone arrays or video calibration and automatically adapts to specific traffic scenes. These specialized detectors are more accurate and more compact than general classifiers, which allows for light-weight usage in low-cost and portable embedded systems. Hence, we implemented our system on an off-the-shelf embedded platform. In the experimental part, we show that the proposed method is able to cover the desired task and outperforms single-cue systems. Furthermore, our co-training framework minimizes the labeling effort without degrading the overall system performance.

This work has been supported by the Austrian FFG projects MobiTrick (825840) and EVis (813399) under the FIT-IT programme and by Lakeside Labs GmbH, Klagenfurt, Austria and funding from the European Regional Development Fund, Interreg IV and the Carinthian Economic Promotion Fund KWF under grant KWF-20214/18354/27107.

Audio-Visual Co-Training

Robust Multi-View Boosting with Priors (Saffari, Leistner, Godec, Bischof)

The increasing amount of visual data and the considerable human effort necessary for labeling these data makes learning from unlabeled data very important. Many learning tasks for computer vision problems can be described by multiple views or multiple features. These views can be exploited in order to learn from unlabeled data, a.k.a. ``multi-view learning''. The main objective of many multi-view learning methods is to increase the agreement of the classifiers trained for each view over the unlabeled samples. In this work, we propose a new multi-view boosting algorithm which specifically encodes the uncertainties over the unlabeled samples in terms of the priors to achieve this goal. Instead of ignoring the unlabeled samples during the training phase of each view, we use the different views to provide an aggregated prior which is then used as a regularization term inside a semi-supervised boosting method. Since we target multi-class applications, we first introduce a multi-class boosting algorithm based on maximizing the mutli-class classification margin. Then, we propose our multi-class semi-supervised boosting algorithm which is able to use priors as a regularization component over the unlabeled data. Since the priors may contain a significant amount of noise, we introduce a new loss function for the unlabeled regularization which is robust to noisy priors. Experimentally, we show that the multi-class boosting algorithms achieves state-of-the-art results in machine learning benchmarks. We also show that the new proposed loss function is more robust compared to other alternatives. Finally, we demonstrate the advantages of our multi-view boosting approach for object category recognition and visual object tracking tasks, compared to other multi-view learning methods.

This work has been supported by the Austrian FFG project MobiTrick (825840) and Outlier (820923) under the FIT-IT program.

Robust Multi-View Boosting with Priors

On-line Multi-class LPBoost (Saffari, Godec, Pock, Leistner, Bischof)

Online boosting is one of the most successful on-line learning algorithms in computer vision. While many challenging online learning problems are inherently multi-class, online boosting and its variants are only able to solve binary tasks. In this paper, we present Online Multi-Class LPBoost (OMCLP) which is directly applicable to multi-class problems. From a theoretical point of view, our algorithm tries to maximize the multi-class soft-margin of the samples. In order to solve the LP problem in online settings, we perform an efficient variant of online convex programming, which is based on primal-dual gradient descent-ascent update strategies. We conduct an extensive set of experiments over machine learning benchmark datasets, as well as, on Caltech101 category recognition dataset. We show that our method is able to outperform other online multi-class methods. We also apply our method to tracking where, we present an intuitive way to convert the binary tracking by detection problem to a multi-class problem where background patterns which are similar to the target class, become virtual classes. Applying our novel model, we outperform or achieve the state-of-the-art results on benchmark tracking videos.

This work has been supported by the Austrian FFG project EVis (813399) and Outlier (820923) under the FIT-IT program and the Austrian Science Fund (FWF) under the doctoral program Confluence of Vision and Graphics W1209.

On-line Multi-class LPBoost

On-line Multi-class LPBoost (Poster)

Context-driven Clustering (Godec, Sternig, Roth, Bischof)

Tracking and detection of objects often require to apply complex models to cope with the large intra-class variability of the foreground as well as the background class. In this work, we reduce the complexity of a binary classification problem by a context-driven approach. The main idea is to use a hidden multi-class representation to capture multi-modalities in the data finally providing a binary classifier. We introduce virtual classes generated by a context-driven clustering, which are updated using an active learning strategy. By further using an on-line learner the classifier can easily be adapted to changing environmental conditions. Moreover, by adding additional virtual classes more complex scenarios can be handled. We demonstrate the approach for tracking as well as detection on different scenarios reaching state-of-the-art results.

This work was supported by the FFG project EVis under the FIT-IT program, the FFG project HIMONI under the COMET programme in co-operation with FTW, the FFG project SECRECT under the Austrian Security Research Programme KIRAS, and the Austrian Science Fund (FWF) under the doctoral program Confluence of Vision and Graphics W1209.

Context-driven Clustering (PDF)

Context-driven Clustering (Presentation)

Context-driven Clustering (Talk)

TransientBoost (Sternig, Godec, Roth, Bischof)

For on-line learning algorithms, which are applied in many vision tasks such as detection or tracking, robust integration of unlabeled samples is a crucial point. Various strategies such as self-training, semi-supervised learning and multiple-instance learning have been proposed. However, these methods are either too adaptive, which causes drifting, or biased by a prior, which hinders incorporation of new (orthogonal) information. Therefore, we propose a new boosting-based on-line learning algorithm (TransientBoost), which is highly adaptive but still robust. This is realized by using an internal multi-class representation and modeling reliable and unreliable data in separate classes. Unreliable data is considered transient, hence we use highly adaptive learning parameters to adapt to fast changes in the scene while errors fade out fast. In contrast, the reliable data is preserved completely and not harmed by wrong updates. We demonstrate our algorithm on two different representative tasks, \ie, object detection and object tracking showing that we can handle typical problems considerable better than existing approaches. In addition, to demonstrate the stability and the robustness, we show long-term experiments for both tasks.

This work was supported by the FFG project EVis under the FIT-IT program, the FFG project HIMONI under the COMET programme in co-operation with FTW, the FFG project SECRECT under the Austrian Security Research Programme KIRAS, and the Austrian Science Fund (FWF) under the doctoral program Confluence of Vision and Graphics W1209.

TransientBoost (PDF)


Longterm Tracking Results

On-line Random Naive Bayes Learning (Godec, Leistner, Saffari, Bischof)

Randomized learning methods (i.e., Forests or Ferns) have shown excellent capabilities for various computer vision applications. However, it was shown that the tree structure in Forests can be replaced by even simpler structures, e.g., Random Naive Bayes classifiers, yielding similar performance. The goal of this paper is to benefit from these findings to develop an efficient on-line learner. Based on the principals of on-line Random Forests, we adapt the Random Naive Bayes classifier to the on-line domain. For that purpose, we propose to use on-line histograms as weak learners, which yield much better performance than simple decision stumps. Experimentally we show, that the approach is applicable to incremental learning on machine learning datasets. Additionally, we propose to use an iir filtering-like forgetting function for the weak learners to enable adaptivity and evaluate our classifier on the task of tracking by detection.

This work was supported by the FFG projects EVis (813399) and OUTLIER (820923) under the FIT-IT programme.

On-line Random Naive Bayes for Tracking

Speeding up Semi-supervised On-line Boosting for Tracking (Godec, Grabner, Leistner, Bischof)

ICVSS Poster Recently, object tracking by detection using adaptive on-line classifiers has been investigated. In this case, the tracking problem is reduced to the discrimination of the current object view from the local background. However, on-line learning may introduce errors, which causes drifting and let the tracker fail. This can be avoided by using semi-supervised on-line learning (i.e., the use of labeled and unlabeled training samples), which allows to limit the drifting problem while still staying adaptive to appearance changes. In particular, this paper extends semi-supervised on-line boosting by a particle filter to achieve a higher frame-rate. Furthermore, a more sophisticated search-space sampling, and an improved update sample selection have been added.

This work has been presented at the AAPR Workshop 2009 in Stainz and at International Computer Vision Summer School 2009 in Sicily.

Speeding up Semi-supervised On-line Boosting for Tracking (PDF)

Speeding up Semi-supervised On-line Boosting for Tracking (Presentation)

Robust Object Tracking using Semi-Supervised Online Boosting (Thesis)

Publications

2013

  1. Hough-based Tracking on Deformable Objects (bib)Martin Godec, Peter M. Roth, and Horst Bischof In Decision Forest, pages 159-173, Springer, 2013
  2. Hough-based Tracking of Non-rigid Objects (bib)Martin Godec, Peter M. Roth, and Horst Bischof Computer Vision and Image Understanding (CVIU) 117 (10): 1245-1256, 2013
  3. Segmentation-based Tracking by Support Fusion (bib)Markus Heber, Martin Godec, Matthias RĂ¼ther, Peter M. Roth, and Horst Bischof Computer Vision and Image Understanding (CVIU) 117 (6): 573-586, 2013

2011

  1. Hough-based Tracking of Non-rigid Objects (bib)Martin Godec, Peter M. Roth, and Horst Bischof In Proc. International Conference on Computer Vision (ICCV), 2011
  2. Improving Classifiers with Unlabeled Weakly-Related Videos (bib)Christian Leistner, Martin Godec, Samuel Schulter, Amir Saffari, Manuel Werlberger, and Horst Bischof In Proc. IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2011
  3. Proc. 16th Computer Vision Winter Workshop (bib)Andreas Wendel, Sabine Sternig, and Martin Godec, eds. Verlag der Technische Universitaet Graz, 2011

2010

  1. Autonomous Audio-Supported Learning of Visual Classifiers for Traffic Monitoring (bib)Horst Bischof, Martin Godec, Christian Leistner, Andreas Starzacher, and Bernhard Rinner IEEE Intelligent Systems, 2010
  2. On-line Random Naive Bayes for Tracking (bib)Martin Godec, Christian Leistner, Amir Saffari, and Horst Bischof In Proc. Int. Conf. on Pattern Recognition, 2010
  3. Context-driven Clustering by Multi-class Classification in an Active Learning Framework (bib)Martin Godec, Sabine Sternig, Peter M. Roth, and Horst Bischof In Proc. Workshop on Use of Context in Video Processing (CVPR), 2010
  4. Audio-Visual Co-Training for Vehicle Classification (bib)Martin Godec, Christian Leistner, Horst Bischof, Andreas Starzacher, and Bernhard Rinner In Proc. IEEE Int'l Conf. on Advanced Video and Signal-Based Surveillance, 2010
  5. Online Multi-View Forests for Tracking (bib)Christian Leistner, Martin Godec, Amir Saffari, and Horst Bischof In Proc. DAGM Symposium, 2010
  6. Online Multi-Class LPBoost (bib)Amir Saffari, Martin Godec, Thomas Pock, Christian Leistner, and Horst Bischof In Proc. IEEE Conf. on Computer Vision and Pattern Recognition, 2010
  7. Robust Multi-View Boosting with Priors (bib)Amir Saffari, Christian Leistner, Martin Godec,, and Horst Bischof In Proc. European Conf. on Computer Vision, 2010
  8. TransientBoost: On-line Boosting with Transient Data (bib)Sabine Sternig, Martin Godec, Peter M. Roth, and Horst Bischof In Proc. IEEE Online Learning for Computer Vision Workshop (CVPR), 2010
  9. Audio-Visual Co-Training for Vehicle Classification (bib)M. Godec, C. Leistner, H. Bischof, A. Starzacher, and B. Rinner In IEEE Conference on Advanced Video and Signal Based Surveillance, 2010

2009

  1. Speeding Up Semi-Supervised On-line Boosting for Tracking (bib)Martin Godec, Helmut Grabner, Christian Leistner, and Horst Bischof In Proc. Workshop of the Austrian Association for Pattern Recognition, 2009
  2. On-line Random Forests (bib)Amir Saffari, Christian Leistner, Jakob Santner, Martin Godec, and Horst Bischof In Proc. IEEE On-line Learning for Computer Vision Workshop, 2009

2013

  1. Tracking-by-Detection using Randomized Online Ensemble Methods (bib)Martin Godec Ph.D. Thesis, Graz University of Technology, Faculty of Computer Science, 2013

2008

  1. Robust Object Tracking using Semi-Supervised On-line Boosting (bib)Martin Godec MSc. Thesis, Graz University of Technology, Faculty of Computer Science, 2008


Copyright 2010 ICG