Natural Image Statistics We us the family of Lp-spherically and Lp-nested symmetric distributions to obtain more accurate morels of natural images and quantitatively assess normative hypotheses about the role of orientation selectivity and divisive normalization in the primate early visual system.

Further Reading

• Sinz F., M. Bethge: Lp-nested symmetric distributions
• Sinz F., E. Simoncelli, M. Bethge: Hierarchical Modeling of Local Image Features through Lp-Nested Symmetric Distributions
• Eichhorn, J. ,F. Sinz, M. Bethge: Natural Image Coding in V1: How Much Use is Orientation Selectivity?
• Sinz F., M. Bethge: The Conjoint Effect of Divisive Normalization and Orientation Selectivity on Redundancy Reduction
• Sinz F., S.Gerwinn, M. Bethge: Characterization of the p-Generalized Normal Distribution

UNIVERSVM: A SVM Implementation for Large Scale Transduction and Inference with a Universum The UNIVERSVM is a SVM implementation written in C++. Its functionality comprises large scale transduction (as described in Large Scale Transductive SVMs), sparse solutions (as described in Trading Convexity for Scalability) and inference with a universum (as described in Inference with the Universum).

• Read online help
• You can download Source code (C++ implementation) at mloss.org (just search for UniverSVM).
  

NEC-Labs ABCDetc letter/digit/symbol dataset (in collaboration Ronan Collobert and Jason Weston at NEC Labs America and Seyda Ertekin at PennState University)

We started to collect a dataset consisting of digits, symbols, uppercase and lowercase letters. At the moment it comprises around 50.000 examples.

If you want to contribute, please download the template , fill out the columns as indicated, scan it at 300 dpi and email it to me. (download page)

Large Scale Optimization (in collaboration with Ronan Collobert, Jason Weston and Leon Bottou at NEC Labs America)

We show how the Concave-Convex Procedure can be applied to Transductive SVMs, which traditionally requires solving a combinatorial search problem. This provides for the first time a highly scalable algorithm in the nonlinear case.

(project details).

Learning Depth from Stereo (Student Research Project) The depth of a point in space can be estimated by observing its image position from two different viewpoints. The classical approach to stereo vision calculates depth from the two projection equations which together form a stereocamera model. An unavoidable preparatory work for this approach is to estimate the parameters of the camera. This can become quite tedious.

In this study, we approached the depth estimation problem from a different point of view by applying generic machine learning algorithms to learn the mapping from image coordinates to spatial position. (project details).