Abstract
Deformable part-based models [1, 2] achieve state-of-the-art performance for ob ject detection, but rely on heuristic initialization dur- ing training due to the optimization of non-convex cost function. This paper investigates limitations of such an initialization and extends earlier methods using additional supervision. We explore strong supervision in terms of annotated ob ject parts and use it to (i) improve model initial- ization, (ii) optimize model structure, and (iii) handle partial occlusions. Our method is able to deal with sub-optimal and incomplete annotations of ob ject parts and is shown to benefit from semi-supervised learning se- tups where part-level annotation is provided for a fraction of positive examples only. Experimental results are reported for the detection of six animal classes in PASCAL VOC 2007 and 2010 datasets. We demon- strate significant improvements in detection performance compared to the LSVM [1] and the Poselet [3] ob ject detectors.