Large margin classifiers for object recognition

In this project, we propose a large margin learning approach for object recognition. Our approach learns to separate two similar images from a different one as much as possible. Our classifier achieves state-of-the-art performance on Catech-101, a standard benchmark dataset.

Left. Averaged pictures of Caltech-101 objects (Attributed to Antonio Torralba). Right. Performance of our large margin classifier (the top line of the connected-black-dot.)

Related publications