For a Human-Centered AI

The Marr Prize goes to FBK

January 4, 2016

The Marr prize, awarded every two years to the best scientific article on artificial vision, was won in the last edition by Trento-based Fondazione Bruno Kessler for a study conducted in collaboration with Microsoft and Carnegie Mellon University.


The work was considered the number one among the 1,700 proposed by research centers around the world at the most important international conference in the field, the “IEEE International Conference on Computer VisionSamuel Rota Bulò, researcher with the TeV (Technologies of Vision) Unit  at Trento-based Fondazione Bruno Kessler’s ICT Center, directed by Paolo Traverso.

With this innovative work, called “Deep Neural Decision Forests”, researchers were able to blend the two techniques used in the field of automatic classification of images (“deep network” and “random forest”) and to get a new system that has produced the best performance ever on the database used in the artificial vision scientific community.

The FBK TeV research unit, currently led by Stefano Messelodi, is no stranger to international first class results.

A study conducted by researchers Oswald Lanz and Elisa Ricci with the TeV Unit, in collaboration with the University of Trento, beat the international competition when it came in first among the works presented during ACMMM 2015, the most important world conference on multimedia, last October in Brisbane, Australia. In this case, the authors had presented a method that enables a more accurate interpretation of the behavior and interaction between people in social events. The scientific study proposed a system to efficiently integrate the information extracted from the video footage of a scene with those sent by sensors worn by people in the scene itself.

The TeV Unit has also just received European funding for “Replicate”, a project in the field of augmented reality that will start in 2016 and which will be coordinated precisely by Trento-based Fondazione Bruno Kessler.


The author/s