My research is within the fascinating domain of Computer Vision. My ultimate research goal is action understanding. The goal is to be able to process visual input (images or videos) in order to answer questions of the type: "who is doing what to whom?".

On my way to achieving this goal I made a little detour into the task of unsupervised learning of visual objects. The algorithms we have developed allow automatic discovery of object categories (and sub-categories) in images in both parametric "top down" and non-parametric "bottom up" manner. Unsupervised category discovery is an important tool for both action understanding and action guided learning. For example, it may allow the action understanding system to automatically discover the object types being interacted by the subjects performing the actions. In addition, the proposed algorithms may be extended to the unsupervised discovery of action categories and sub-categories.

My previous project was a non-parametric system capable of detecting highly mobile parts of non-rigid objects, such as human hands, horse hooves, etc. The topic of our current project is showing how using this system as a human body parts detector naturally facilitates the task of action understanding.