picture of Radu Timofte

Dr. Radu Timofte

- Research Group Leader -

Computer Vision Laboratory
Sternwartstrasse 7
ETH Zentrum
CH - 8092 Zürich, Switzerland

Office: ETF C107
Tel: +41 44 63 25279
Fax: +41 44 63 21199

Radu.Timofte at vision.ee.ethz.ch


Snowbird, WACV2009

Radu Timofte, Takeo Kanade, Reyes Rios-Cabrera, Maria Pavlou, Agustin Alberto Ortega Jimenez

RESEARCH INTERESTS
Multi-Class Multi-View Object Detection, Recognition, Segmentation, Tracking
Sparse and Collaborative Representations
Machine Learning
Artificial Intelligence
SOME RESEARCH TOPICS
Multi-view traffic sign detection, recognition, and 3D localisation
Traffic sign 3D mapping
Input videos coming from several cameras mounted on a van are processed offline for extracting the traffic signs and their 3D mapped positions. We focus on the 2D and 3D analysis by proposing a fast optimal segmentation followed by AdaBoost cascades for detection, hierarchies of SVM for validation/recognition, 3D hypotheses reconstruction and MDL formulation for best selection of the subset of hypotheses as solution to our problem.
Integrating Object Detection With 3D Tracking Towards A Better Driver Assistance System
DAS sample
An input video must be processed in real-time for driver assistance purposes. We focus on achieving real-time performance on frame-level detection and recognition, as well in 3D pose tracking of the detected traffic signs. In this way we have not only the detected traffic signs facing the car but also their relative orientation, their 3D poses.
Hough Transform and 3D SURF for Robust Three Dimensional Classification
3D features
Inspired by 2D methods, recently researchers have started to work with local features. In keeping with this strand, we propose a new robust 3D shape classification method. It contains two main contributions. First, we extend a robust 2D feature descriptor, SURF, to be used in the context of 3D shapes. Second, we show how 3D shape class recognition can be improved by probabilistic Hough transform based methods, already popular in 2D. Through our experiments on partial shape retrieval, we show the power of the proposed 3D features. Their combination with the Hough transform yields superior results for class recognition on standard datasets. The potential for the applicability of such a method in classifying 3D obtained from Structure-from-Motion methods is promising, as we show in some initial experiments.
Four Color Theorem for Fast Early Vision
3D features
Recent work on early vision such as image segmentation, image restoration, stereo matching, and optical flow models these problems using Markov Random Fields. Although this formulation yields an NP-hard energy minimization problem, good heuristics have been developed based on graph cuts and belief propagation. Nevertheless both approaches still require tens of seconds to solve stereo problems on recent PCs. Such running times are impractical for optical flow and many image segmentation and restoration problems. We show how to reduce the computational complexity of belief propagation by applying the Four Color Theorem to limit the maximum number of labels in the underlying image segmentation to at most four. We show that this provides substantial speed improvements for large inputs to a variety of vision problems, while maintaining competitive result quality.
Multi-view Manhole 3D Mapping
manhole detection
Given as input videos recorded from a van, detect accurately all the manholes in short distance from the camera. Fitting a ground plane under the cars' wheels is beneficial for working on ground projected images, where the manholes have a normalized view. The pipeline is useful for accurate 3D mapping of manholes when the camera calibrations/poses are known, obtained offline through a bundle adjustment procedure. Another usage is for introducing accurate landmarks for bundle adjustment, assuming that a map of the manholes is available.
KUL Belgium Traffic Signs Dataset
Traffic sign samples
BelgiumTS is a large dataset with 10000+ traffic sign annotations, thousands of physically distinct traffic signs. 4 video sequences recorded with 8 high resolution cameras mounted on a van, summing more than 3 hours, with traffic sign annotations, camera calibrations and poses. About 16000 background images. The material is captured in Belgium, in urban environments from Flanders region, by GeoAutomation.
KUL Belgium Traffic Sign Classification Benchmark
Traffic sign samples
BelgiumTSC is built for traffic sign classification purposes. Is is a subset of BelgiumTS dataset and contains cropped images around annotations for 62 different classes of traffic signs. BelgiumTSC is split in a training part with 4591 images and a testing part with 2534 images. The split follows the split from BelgiumTS. On average there are 3 images/annotations for each physically distinct traffic sign. The material is recorded in Belgium, in urban environments from Flanders region, by GeoAutomation.

NTIRE 2016: New Trends in Image Restoration and Enhancement, workshop in conjunction with ACCV 2016, Taiwan.


DEMO: Prediction of attractiveness, age, and gender from a profile image.


Publications

[53] R. Rothe, R. Timofte, and L. Van Gool. Deep expectation of real and apparent age from a single image without facial landmarks. Journal International Journal of Computer Vision (IJCV 2016), 2016. [ bib |  pdf |  Models and IMDB-WIKI dataset |  DEMO: howhot.io ]
 
[52] R. Timofte and L. Van Gool. Anchored Fusion for Image Restoration. In 23rd International Conference on Pattern Recognition (ICPR 2016), December 2016, Mexico. [ bib | pdf |  Source codes and images (soon) ]
 
[51] E. Agustsson, R. Timofte, and L. Van Gool. Regressor Basis Learning for Anchored Super-Resolution. In 23rd International Conference on Pattern Recognition (ICPR 2016), December 2016, Mexico. [ bib | pdf |  Source codes and images (soon) ]
 
[50] E. Agustsson, R. Timofte, and L. Van Gool. k2-means for fast and accurate large scale clustering. In arXiv:1605.09299, May 2016. [ bib |  pdf |  Source codes and data (soon) ]
 
[49] J. Wu, R. Timofte, and L. Van Gool. Demosaicing based on Directional Difference Regression and Efficient Regression Priors. Journal IEEE Transactions on Image Processing (TIP 2016), 2016. [ bib |  pdf |  Source codes and images ]
 
[48] A. Volokitin, R. Timofte, and L. Van Gool. Deep Features or Not: Temperature and Time Prediction in Outdoor Scenes. In Robust Features for Computer Vision workshop (CVPR 2016), June 2016, US. [ bib |  pdf |  Source codes and images ]
 
[47] M. Uricar, R. Timofte, R. Rothe, J. Matas, and L. Van Gool. Structured Output SVM Prediction of Apparent Age, Gender and Smile From Deep Features. In ChaLearn Looking at People and Faces of the World: Face Analysis Workshop and Challenge (CVPR 2016), June 2016, US. (3rd place of LAP challenge on apparent age estimation) [ bib |  pdf |  Source codes and models ]
 
[46] T. Kroeger, R. Timofte, D. Dai, and L. Van Gool. Fast Optical Flow using Dense Inverse Search. In European Conference on Computer Vision (ECCV 2016), October 2016, Netherlands. [ bib |  pdf |  Project page ]
 
[46] T. Kroeger, R. Timofte, D. Dai, and L. Van Gool. Fast Optical Flow using Dense Inverse Search. In arXiv:1603.03590, March 2016. [ bib |  pdf |  Project page ]
 
[45] R. Timofte, J. Kwon, and L. Van Gool. PICASO: PIxel Correspondences And SOft Match Selection for Real-time Tracking. Journal Computer Vision and Image Understanding (CVIU 2016), 2016. [ bib |  pdf |  Source codes and images (soon) ]
 
[44] S. Manen, R. Timofte, D. Dai, and L. Van Gool. Leveraging single for multi-target tracking using a novel trajectory overlap affinity measure. In IEEE Winter Conference on Applications of Computer Vision (WACV 2016), March 2016, US. [ bib |  pdf |  Suppl. material ]
 
[43] R. Timofte, R. Rothe, and L. Van Gool. Seven ways to improve example-based single image super resolution. In IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR 2016), June 2016, US. [ bib |  pdf |  Supplementary material |  L20 dataset ]
 
[43] R. Timofte, R. Rothe, and L. Van Gool. Seven ways to improve example-based single image super resolution. In arXiv:1511.02228, November 2015. [ bib |  pdf ]
 
[42] R. Rothe, R. Timofte, and L. Van Gool. Some like it hot - visual guidance for preference prediction. In 2016 IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR 2016), June 2016, US. [ bib |  pdf |  DEMO: howhot.io ]
 
[42] R. Rothe, R. Timofte, and L. Van Gool. Some like it hot - visual guidance for preference prediction. In arXiv:1510.07867, October 2015. [ bib |  pdf |  DEMO: howhot.io ]
 
[41] R. Rothe, R. Timofte, and L. Van Gool. DEX: Deep EXpectation of apparent age from a single image. In Looking at People Workshop at International Conference on Computer Vision (ICCV 2015), December 2015, Chile. (Winner of LAP challenge on apparent age estimation) (NVIDIA ChaLearn LAP 2015 Best Paper Award) [ bib |  pdf |  Models and IMDB-WIKI dataset |  DEMO: howhot.io ]
 
[40] R. Rothe, R. Timofte, and L. Van Gool. DLDR: Deep Linear Discriminative Retrieval for cultural event classification from a single image. In Looking at People Workshop at International Conference on Computer Vision (ICCV 2015), December 2015, Chile. (Top entry in LAP challenge on cultural event recognition) [ bib |  pdf ]
 
[39] R. Timofte*, V. De Smet*, and L. Van Gool. Semantic super-resolution: when and where is it useful? Journal Computer Vision and Image Understanding (CVIU 2015), 2015. (* equal contributions) [ bib |  pdf ]
 
[38] R. Rothe, R. Timofte, and L. Van Gool. Efficient Regression Priors for Reducing Image Compression Artifacts. In 22nd IEEE International Conference on Image Processing (ICIP 2015), September 2015, Canada. [ bib |  pdf |  Source codes and images ]
 
[37] J. Wu, R. Timofte, and L. Van Gool. Efficient Regression Priors for Post-processing Demosaiced Images. In 22nd IEEE International Conference on Image Processing (ICIP 2015), September 2015, Canada. [ bib |  pdf |  Source codes and images (soon) ]
 
[36] H. Honda, R. Timofte, and L. Van Gool. Make My Day - High-Fidelity Color Denoising with Near-Infrared. In 11th IEEE Workshop on Perception Beyond the Visible Spectrum (PBVS) - CVPR Workshops, June 2015, USA. [ bib |  pdf |   MMD images (RGB+NIR) ]
 
[35] D. Dai, T. Kroeger, R. Timofte, and L. Van Gool. Metric Imitation by manifold transfer for efficient vision applications. In 2015 IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR 2015), June 2015, USA. [ bib |  pdf |  Project ]
 
[34] D. Dai, R. Timofte, and L. Van Gool. Jointly Optimized Regressors for Image Super-resolution. Journal Computer Graphics Forum, vol. 34, num. 2, pp. 95-104, June 2015. [ bib |  pdf |  Source codes ]
 
[34] D. Dai, R. Timofte, and L. Van Gool. Jointly Optimized Regressors for Image Super-resolution. In The 36th Annual Conference of the European Association for Computer Graphics (EUROGRAPHICS 2015), May 2015, Switzerland. [ bib |  pdf |  Source codes ]
 
[33] R. Timofte and L. Van Gool. SparseFlow: Sparse Matching for Small to Large Displacement Optical Flow. In IEEE Winter Conference on Applications of Computer Vision (WACV 2015), January 2015, US. [ bib |  pdf |  Source codes ]
 
[32] J. Wu, R. Timofte, and L. Van Gool. Learned Collaborative Representations for Image Classification. In IEEE Winter Conference on Applications of Computer Vision (WACV 2015), January 2015, US. [ bib |  pdf |  Source codes ]
 
[31] T. Kroeger, D. Dai, R. Timofte, and L. Van Gool. Discovery of Sets of Mutually Orthogonal Vanishing Points in Videos. In 1st Workshop on Benchmarking multi-target tracking (BMTT @ WACV 2015), January 2015, US. [ bib |  pdf | more ]
 
[30] R. Timofte, V. De Smet, and L. Van Gool. A+: Adjusted Anchored Neighborhood Regression for Fast Super-Resolution. In Asian Conference on Computer Vision (ACCV 2014), November 2014, Singapore. [ bib |  pdf |  Supplementary material |  Source codes ]
 
[29] R. Timofte and L. Van Gool. Iterative Nearest Neighbors. Journal Pattern Recognition (PR 2014), 2014. [ bib |  pdf |  Source codes ]
 
[28] M. Pedersoli, R. Timofte, T. Tuytelaars, and L. Van Gool. An Elastic Deformation Field Model for Object Detection and Tracking. In International Journal of Computer Vision, June 2014. [ bib |  pdf ]
 
[27] M. Pedersoli, R. Timofte, T. Tuytelaars, and L. Van Gool. An Elastic Deformation Field Model for Object Detection and Tracking. In Tech. Report, KU Leuven/ESAT/PSI/1402, April 2014, Belgium. [ bib |  pdf ]
 
[26] B. Verhagen, R. Timofte, and L. Van Gool. Scale-Invariant Line Descriptors for Wide Baseline Matching. In Winter Conference on Applications of Computer Vision (WACV 2014), March 2014, USA. [ bib |  pdf |  Source codes ]
 
[25] R. Timofte, V. De Smet, and L. Van Gool. Anchored Neighborhood Regression for Fast Example-Based Super-Resolution. In International Conference on Computer Vision (ICCV 2013), December 2013, Australia. [ bib |  pdf |  poster |  Supplementary material |  Source codes ]
 
[24] M. Mathias, R. Benenson, R. Timofte, and L. Van Gool. Handling Occlusions with Franken-classifiers. In International Conference on Computer Vision (ICCV 2013), December 2013, Australia. [ bib | pdf (includes supplementary material) ]
 
[23] R. Timofte and L. Van Gool. Adaptive and Weighted Collaborative Representations for Image Classification. Journal Pattern Recognition Letters (PRL 2014), vol.43, July 2014. [ bib | pdf |  Source codes ]
 
[22] F.  Schouwenaars, R. Timofte, and L. Van Gool. Robust Scene Stitching in Large Scale Mobile Mapping. In British Machine Vision Conference (BMVC 2013), September 2013, UK. [ bib | pdf ]
 
[21] R. Timofte. Sparse and Collaborative Representations for Computer Vision. PhD Thesis (supervisor: Prof. Dr. Ing. Luc Van Gool), KU Leuven, June 2013, Belgium. [ bib | pdf ]
 
[20] R. Timofte, and L. Van Gool. Efficient Loopy Belief Propagation using the Four Color Theorem. In Advanced Topics in Computer Vision, Series Title: Advances in Computer Vision and Pattern Recognition . Editors: G. Farinella, S. Battiato, and R. Cipolla. Springer-Verlag London Ltd., 2013. (in print) [ bib | pdf ]
 
[19] M.  Mathias*, R. Timofte*, R. Benenson, and L. Van Gool. Traffic Sign Recognition - How far are we from the solution?. In International Joint Conference on Neural Networks (IJCNN 2013), August 2013, Dallas, USA. (* equal contributions) (Winner of GTSDB challenge on traffic sign detection) (Third entry of GTSRB challenge on traffic sign classification) [ bib |  html | pdf ]
 
[18] R. Timofte and L. Van Gool. Weighted Collaborative Representation and Classification of Images. In 21st International Conference on Pattern Recognition (ICPR 2012), November 2012, Japan. (Best Scientific Paper Award) [ bib | pdf ]
 
[17] R. Timofte, T. Tuytelaars, and L. Van Gool. Naive Bayes Image Classification: beyond Nearest Neighbors. In Asian Conference on Computer Vision (ACCV 2012), November 2012, Korea. [ bib | pdf ]
 
[16] R. Timofte and L. Van Gool. Automatic Stave Discovery for Musical Facsimiles. In Asian Conference on Computer Vision (ACCV 2012), November 2012, Korea. [ bib |  Supplementary material | pdf ]
 
[15] R. Benenson, M. Mathias, R. Timofte, and L. Van Gool. Fast Stixel Computation for Fast Pedestrian Detection. In 3rd IEEE Workshop on Computer Vision in Vehicle Technology: From Earth to Mars (CVVT @ ECCV 2012), October 2012, Italy. (Best Paper Award) [ bib | pdf | more ]
 
[14] B. Günyel, R. Benenson, R. Timofte, and L. Van Gool. Stixels Motion Estimation without Optical Flow Computation. In 12th European Conference on Computer Vision (ECCV 2012), October 2012, Italy. [ bib | pdf | more ]
 
[13] R. Timofte and L. Van Gool. A Training-free Classification Framework for Textures, Writers, and Materials. In British Machine Vision Conference (BMVC 2012), September 2012, UK. [ bib | pdf ]
 
[12] R. Timofte and L. Van Gool. Iterative Nearest Neighbors for Classification and Dimensionality Reduction. In 2012 IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR 2012), June 2012, USA. [ bib | pdf | INNC.m ]
 
[11] R. Benenson, M. Mathias, R. Timofte, and L. Van Gool. Pedestrian Detection at 100 Frames per Second. In 2012 IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR 2012), June 2012, USA. [ bib | pdf | more ]
 
[10] R. Timofte, K. Zimmermann, and L. Van Gool. Multi-view Traffic Sign Detection, Recognition, and 3D Localisation. Journal of Machine Vision and Applications (MVA 2011), DOI 10.1007/s00138-011-0391-3, December 2011, Springer-Verlag. [ bib |  html | pdf ]
 
[9] V. Lasdas, R. Timofte, and L. Van Gool. Non-Parametric Motion-Priors for Flow Understanding. In IEEE Workshop on Applications of Computer Vision (WACV 2012), January 2012, USA. [ bib |  children_sequence.zip | pdf ]
 
[8] R. Timofte, V.A. Prisacariu, L.J. Van Gool, and I. Reid. Combining Traffic Sign Detection with 3D Tracking Towards Better Driver Assistance. Emerging Topics in Computer Vision and its Applications (editor: C.H. Chen). World Scientific Publishing. September 2011. [ bib |  html | pdf ]
 
[7] R. Timofte and L. Van Gool. Multi-view Manhole Detection, Recognition, and 3D Localisation. In 1st IEEE/ISPRS Workshop on Computer Vision for Remote Sensing of the Environment (CVRS @ ICCV 2011), November 2011, Barcelona, Spain. [ bib | pdf ]
 
[6] R. Benenson, R. Timofte, and L. Van Gool. Stixels Estimation Without Depth Map Computation. In 2nd IEEE Workshop on Computer Vision in Vehicle Technology: From Earth to Mars (CVVT @ ICCV 2011), November 2011, Barcelona, Spain. [ bib | pdf | more ]
 
[5] R. Timofte and L. Van Gool. Sparse Representation Based Projections. In British Machine Vision Conference (BMVC 2011), 2011, UK. [ bib | pdf ]
 
[4] R. Timofte and L. Van Gool. Four Colour Theorem for Fast Early Vision. In Asian Conference on Computer Vision (ACCV 2010), November 2010, New Zealand. [ bib |  Supplementary material | pdf ]
 
[3] J. Knopp, M. Prasad, G. Willems, R. Timofte and L. Van Gool. Hough Transform and 3D SURF for Robust Three Dimensional Classification. In European Conference on Computer Vision (ECCV 2010), September 2010, Greece. [ bib |  html | pdf ]
 
[2] V.A. Prisacariu, R. Timofte, K. Zimmermann, I. Reid, and L. Van Gool. Integrating Object Detection with 3D Tracking Towards a Better Driver Assistance System. In International Conference on Pattern Recognition (ICPR 2010), August 2010, Turkey. [ bib |  html | pdf ]
 
[1] R. Timofte, K. Zimmermann, and L. Van Gool. Multi-view Traffic Sign Detection, Recognition, and 3D Localisation. In IEEE Workshop on Applications of Computer Vision (WACV 2009), December 2009, USA. [ bib |  html | pdf ]
 
Latest change 5:17 PM 2016/08/02