Given a set of images containing objects from the same category, the task ofimage co-localization is to identify and localize each instance. This papershows that this problem can be solved by a simple but intriguing idea, that is,a common object detector can be learnt by making its detection confidencescores distributed like those of a strongly supervised detector. Morespecifically, we observe that given a set of object proposals extracted from animage that contains the object of interest, an accurate strongly supervisedobject detector should give high scores to only a small minority of proposals,and low scores to most of them. Thus, we devise an entropy-based objectivefunction to enforce the above property when learning the common objectdetector. Once the detector is learnt, we resort to a segmentation approach torefine the localization. We show that despite its simplicity, our approachoutperforms state-of-the-art methods.
展开▼
