Arnaud Declercq Homepage

PhD Thesis

My research topic is "Unsupervised Learning of Articulated Visual Feature Hierarchies". The goal is to use videos to incrementally learn complex models of objects. My work is divided in two main parts :

• Using the temporal information to improve the learning process.
• Describing the complex relations (artculated) between features.

DEA's Thesis

My research topic is "On-line Simultaneous Learning and Tracking of Visual Feature Graphs". The idea is base on the fact that a tracking algorithm often relies on a good model of the object to track. On the other side, a learning algorithm needs good tracking conditions to give good results. Here we try to incrementally learn a graphical model of an object using a tracking algorithm supported by this imperfect model. So, for each frame of a video, we apply this two-steps process :

• Using the current model to track the object.
• Improve the learned static graphical model from this tracking result.

DEA's Thesis report: [Pdf]