Beyond sliding windows: Object detection based on hierarchical segmentation model

In this paper, we propose a new selective search strategy for object detection using hierarchical segmentation model. Our method differs from exhaustive search in that the former is class-independent and generates less candidate positions. The experimental results show that this selective search method can recall almost all objects in the five object classes of Caltech 101 dataset using only a few hundred locations per image. Another advantage of the proposed method is that it can go beyond the detection task and achieve good object segmentation.