Scene understanding includes many related sub-tasks, such as scene, depth estimation, object detection, etc. Each of these-tasks is often notoriously hard, and state-of-the-art classifiers already for many of them. These classifiers operate on the same raw image and correlated outputs. It is desirable to have an algorithm that can such correlation without requiring any changes to the inner workings of classifier. We propose Feedback Enabled Cascaded Classification Models (FE-CCM), that optimizes all the sub-tasks, while requiring only a `black-box' to the original classifier for each sub-task. We use a two-layer of classifiers, which are repeated instantiations of the original ones, the output of the first layer fed into the second layer as input. Our method involves a feedback step that allows later classifiers to earlier classifiers information about which error modes to focus on. We that our method significantly improves performance in all the sub-tasks in domain of scene understanding, where we consider depth estimation, scene, event categorization, object detection, geometric labeling and detection. Our method also improves performance in two robotic: an object-grasping robot and an object-finding robot.
No takes yet. Share an insight, caveat, or question.
Li et al. (2011) studied this question.