Abstract

In this paper we introduce a new problem which we call ob- ject co-detection. Given a set of images with ob jects observed from two or multiple images, the goal of co-detection is to detect the ob jects, es- tablish the identity of individual ob ject instance, as well as estimate the viewpoint transformation of corresponding ob ject instances. In designing a co-detector, we follow the intuition that an ob ject has consistent ap- pearance when observed from the same or different viewpoints. By mod- eling an ob ject using state-of-the-art part-based representations such as [1,2], we measure appearance consistency between ob jects by comparing part appearance and geometry across images. This allows to effectively account for ob ject self-occlusions and viewpoint transformations. Exten- sive experimental evaluation indicates that our co-detector obtains more accurate detection results than if ob jects were to be detected from each image individually. Moreover, we demonstrate the relevance of our co- detection scheme to other recognition problems such as single instance ob ject recognition, wide-baseline matching, and image query.