Computer Vision and Machine Learning

Christoph Lampert

The Computer Vision and Machine Learning group studies and develops algorithms in the area of statistical machine learning and computer vision, for example for the task of natural image understanding. Using tools from probability theorem, optimization and functional analysis, the members of the group study questions such as: How can a computer identify which objects are visible in an image? and How can a system learn continuously over a long time in a changing environment.

Contact
Christoph Lampert
Institute of Science and Technology Austria (IST Austria)
Am Campus 1
A – 3400 Klosterneuburg

Phone: +43 (0)2243 9000-3101
E-mail: christoph.lampert@ist.ac.at

CV

Publication list

Lampert Group website

Assistant
Elisabeth Hacker

Phone: +43 (0)2243 9000-1015
E-mail: elisabeth.hacker@remove-this.ist.ac.at

Team

Current Projects

  • Lifelong learning for visual scene understanding (L3ViSU)
    Our goal in the L3ViSu project is to develop and analyse algorithms that use continuous, open-ended machine learning from visual input data (images and videos) in order to enable a computer to interpret visual scenes. The main underlying hypothesis is that we can only significantly improve the state of the art in computer vision algorithms by giving them access to background and contextual knowledge about the visual world, and that the most effective way to obtain such knowledge is by extracting it (semi-)automatically from incoming visual stimuli. Consequently, at the core of L3ViSU lies the idea of life-long visual learning, which we study both on the level of machine learning theory as well as application. See http://www.ist.ac.at/~chl/erc
  • Efficient object and action localization
    The tasks of identifying and locating objects in natural images are crucial components for any automatic image understanding system, and they are inherently linked. We work on developing fast learning and inference techniques for object localization based on formulating the problem as a regression task between structured spaces instead of a sequence of classification tasks. This allows the use of global optimization techniques, such as branch-and-bound, instead of the exhaustive search used by currently dominating sliding window approaches.
  • Attribute-enabled representations
    Image understanding on a semantic level requires knowledge not only about the presence and location of objects, but also about their properties. We work on extending existing image and object representations by attributes. These are characteristic properties of a scene or an object category that can be predicted from the image information, but also have a semantic interpretation. Our goal is to show that attribute-enabled representations can achieve higher classification accuracy than purely object-centric ones, that they require less training data and that they allow solving new image-related prediction tasks, for example for making decisions about the importance of objects in a scene.
  • Structured prediction and learning
    We are working on extending and improving existing techniques for the prediction of structured objects using machine learning techniques, in particular graph labeling problems. Treating these in a framework of empirical risk minimization allows us to develop algorithms that can learn prediction functions that are tailored to the specific prediction tasks and thereby achieve higher prediction accuracy. In particular, we work on developing techniques for handling the cases of ambiguous input and outputs.

Selected Publications

  • Lampert C.H, Nickisch H, Harmeling S. 2009. Learning to detect unseen object classes by between-class attribute transfer. IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Miami Beach, USA.
  • Lampert CH, Blaschko MB, Hofmann T. 2008. Beyond sliding windows: Object localization by efficient subwindow search”, IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Anchorage, AK, USA.
  • Blaschko MB, Lampert CH.2008. Learning to localize objects with structured output regression, European Conference on Computer Vision, Marseilles, France.

Career

2010 Assistant Professor, IST Austria
2007–2010 Senior Research Scientist, Max-Planck Institute, Tübingen, Germany
2004–2007 Senior Researcher, German Research Center for Artificial Intelligence, Kaiserslautern, Germany
2003 PhD, University of Bonn, Germany

Selected Distinctions

2013 Elected member of the "Junge Kurie" (Young Academy) of the Austrian Acadamy of Sciences (ÖAW)
2012 ERC Starting Grant
2008 Main Price, German Society for Pattern Recognition (DAGM)
2008 Best Paper Award, IEEE Conference for Computer Vision and Pattern Recognition (CVPR)
2008 Best Student Paper Award, European Conference for Computer Vision (ECCV)

To top