Abstract
We present an approach for identifying a set of candidate ob jects in a given image. This set of candidates can be used for ob ject recognition, segmentation, and other ob ject-based image parsing tasks. To generate the proposals, we identify critical level sets in geodesic dis- tance transforms computed for seeds placed in the image. The seeds are placed by specially trained classifiers that are optimized to discover ob jects. Experiments demonstrate that the presented approach achieves significantly higher accuracy than alternative approaches, at a fraction of the computational cost.