A5007656938- Advanced Image and Video Retrieval Techniques
- Advanced Vision and Imaging
- Video Surveillance and Tracking Methods
- Robotics and Sensor-Based Localization
- Image Retrieval and Classification Techniques
- Advanced Neural Network Applications
- Image Enhancement Techniques
- Image and Object Detection Techniques
- Face and Expression Recognition
- Handwritten Text Recognition Techniques
- Image Processing Techniques and Applications
- Human Pose and Action Recognition
- Face recognition and analysis
- Advanced Image Processing Techniques
- Remote-Sensing Image Classification
- Color Science and Applications
- Anomaly Detection Techniques and Applications
- Optical measurement and interference techniques
- Industrial Vision Systems and Defect Detection
- Domain Adaptation and Few-Shot Learning
- Speech and Audio Processing
- Video Analysis and Summarization
- Multimodal Machine Learning Applications
- Image Processing and 3D Reconstruction
- Visual Attention and Saliency Detection
Czech Technical University in Prague
2016-2025
Tampere University
2017-2019
Center for Economic Research and Graduate Education – Economics Institute
2017-2018
Signal Processing (United States)
2017
GE Global Research (United States)
2012
Delft University of Technology
2011
University of Illinois Urbana-Champaign
2011
Wuhan University
2010
Harvard University
2010
State Key Laboratory of Information Engineering in Surveying Mapping and Remote Sensing
2010
We develop a common theoretical framework for combining classifiers which use distinct pattern representations and show that many existing schemes can be considered as special cases of compound classification where all the are used jointly to make decision. An experimental comparison various classifier combination demonstrates rule developed under most restrictive assumptions-the sum rule-outperforms other combinations schemes. A sensitivity analysis estimation errors is carried out this...
This paper investigates long-term tracking of unknown objects in a video stream. The object is defined by its location and extent single frame. In every frame that follows, the task to determine object's or indicate not present. We propose novel framework (TLD) explicitly decomposes into tracking, learning, detection. tracker follows from detector localizes all appearances have been observed so far corrects if necessary. learning estimates detector's errors updates it avoid these future....
We present DeblurGAN, an end-to-end learned method for motion deblurring. The learning is based on a conditional GAN and the content loss. DeblurGAN achieves state-of-the art performance both in structural similarity measure visual appearance. quality of deblurring model also evaluated novel way real-world problem - object detection (de-)blurred images. 5 times faster than closest competitor Deep-Deblur [25]. introduce generating synthetic blurred images from sharp ones, allowing realistic...
Results of the ICDAR 2015 Robust Reading Competition are presented. A new Challenge 4 on Incidental Scene Text has been added to Challenges Born-Digital Images, Focused Images and Video Text. is run a newly acquired dataset 1,670 images evaluating Localisation, Word Recognition End-to-End pipelines. In addition, for 3 substantially updated with more video sequences accurate ground truth data. Finally, tasks assessing system performance have introduced all Challenges. The competition took...
Abstract The wide-baseline stereo problem, i.e. the problem of establishing correspondences between a pair images taken from different viewpoints is studied. A new set image elements that are put into correspondence, so called extremal regions , introduced. Extremal possess highly desirable properties: closed under (1) continuous (and thus projective) transformation coordinates and (2) monotonic intensities. An efficient (near linear complexity) practically fast detection algorithm frame...
Short-term tracking is an open and challenging problem for which discriminative correlation filters (DCF) have shown excellent performance.We introduce the channel spatial reliability concepts to DCF provide a novel learning algorithm its efficient seamless integration in filter update process. The map adjusts support part of object suitable tracking. This both allows enlarge search region improves non-rectangular objects. Reliability scores reflect channel-wise quality learned are used as...
This paper shows that the performance of a binary classifier can be significantly improved by processing structured unlabeled data, i.e. data are if knowing label one example restricts labeling others. We propose novel paradigm for training from labeled and examples we call P-N learning. The learning process is guided positive (P) negative (N) constraints which restrict set. evaluates on identifies have been classified in contradiction with structural augments set corrected samples an...
A new robust matching method is proposed. The progressive sample consensus (PROSAC) algorithm exploits the linear ordering defined on set of correspondences by a similarity function used in establishing tentative correspondences. Unlike RANSAC, which treats all equally and draws random samples uniformly from full set, PROSAC are drawn progressively larger sets top-ranked Under mild assumption that measure predicts correctness match better than guessing, we show achieves large computational...
An end-to-end real-time scene text localization and recognition method is presented. The performance achieved by posing the character detection problem as an efficient sequential selection from set of Extremal Regions (ERs). ER detector robust to blur, illumination, color texture variation handles low-contrast text. In first classification stage, probability each being a estimated using novel features calculated with O(1) complexity per region tested. Only ERs locally maximal are selected...
This paper proposes a novel method for tracking failure detection. The detection is based on the Forward-Backward error, i.e. performed forward and backward in time discrepancies between these two trajectories are measured. We demonstrate that proposed error enables reliable of failures selection video sequences. approach complementary to commonly used normalized cross-correlation (NCC). Based we propose object tracker called Median Flow. State-of-the-art performance achieved challenging...
The Visual Object Tracking challenge 2015, VOT2015, aims at comparing short-term single-object visual trackers that do not apply pre-learned models of object appearance. Results 62 are presented. number tested makes VOT 2015 the largest benchmark on tracking to date. For each participating tracker, a short description is provided in appendix. Features VOT2015 go beyond its VOT2014 predecessor are: (i) new dataset twice as large with full annotation targets by rotated bounding boxes and...
A computational problem that arises frequently in computer vision is of estimating the parameters a model from data have been contaminated by noise and outliers. More generally, any practical system seeks to estimate quantities noisy measurements must at its core some means dealing with contamination. The random sample consensus (RANSAC) algorithm one most popular tools for robust estimation. Recent years seen an explosion activity this area, leading development number techniques improve...
This paper addresses the problem of single-target tracker performance evaluation. We consider measures, dataset and evaluation system to be most important components propose requirements for each them. The are basis a new methodology that aims at simple easily interpretable comparison. ranking-based equivalence in terms statistical significance practical differences. A fully-annotated with per-frame annotations several visual attributes is introduced. diversity its properties maximized novel...
The Visual Object Tracking challenge VOT2017 is the fifth annual tracker benchmarking activity organized by VOT initiative. Results of 51 trackers are presented; many state-of-the-art published at major computer vision conferences or journals in recent years. evaluation included standard and other popular methodologies a new "real-time" experiment simulating situation where processes images as if provided continuously running sensor. Performance tested typically far exceeds baselines. source...