Abstract
We propose a model for classification and detection of ob ject classes where the number of classes may be large and where multiple in- stances of ob ject classes may be present in an image. The algorithm com- bines a bottom-up, low-level, procedure of a bag-of-words naive Bayes phase for winnowing out unlikely ob ject classes with a high-level proce- dure for detection and classification. The high-level process is a hybrid of a voting method where votes are filtered using beliefs computed by a class-specific graphical model. In that sense, shape is both explicit (determining the voting pattern) and implicit (each ob ject part votes independently) — hence the term ”semi-explicit shape model”.