Incremental learning of object detectors using a visual shape alphabet, by Opelt, Pinz, and Zisserman.
Here is the abstract from the paper:
We address the problem of multiclass object detection. Our aims are to enable models for new categories to ben- efit from the detectors built previously for other categories, and for the complexity of the multiclass system to grow sub-linearly with the number of categories. To this end we intro- duce a visual alphabet representation which can be learnt incrementally, and explicitly shares boundary fragments (contours) and spatial configurations (relation to centroid) across object categories. We develop a learning algorithm with the following novel contributions: (i ) AdaBoost is adapted to learn jointly, based on shape features; (ii) a new learning sched- ule enables incremental additions of new categories; and (iii) the algorithm learns to detect objects (instead of cate- gorizing images). Frthermore, we show that category sim- ilarities can be predicted from the alphabet. We obtain excellent experimental results on a variety of complex categories over several visual aspects. We show that the sharing of shape features not only reduces the num- ber of features required per category, but also often im- proves recognition performance, as compared to individual detectors which are trained on a per-class basis.
http://eprints.pascal-network.org/archive/00002117/01/opelt06a.pdf
No comments:
Post a Comment