Categorization of Multiple Objects in a Scene without Semantic Segmentation

Abstract

In this paper, we present a novel approach for multi-object categorization within the Bag-of-Features (BoF) framework. We integrate a biased sampling component with a multi-instance multi-label leaning and classification algorithm into the categorization system. With the proposed approach, we addresses two issues in BoF related methods simultaneously: how to avoid scene modeling and how to predict labels of an image without explicitly semantic segmentation when multiple categories of objects are co-existing. The experimental results on VOC2007 dataset show that the proposed method outperforms others in the challenge’s classification task and achieves good performance in multi-object categorization tasks.