Efficient Subwindow Search: A Branch and Bound Framework for Object Localization

Abstract

Abstract-Most successful object recognition systems rely on binary classification, deciding only if an object is present or not, but not providing information on the actual object location. To estimate the object's location one can take a sliding window approach, but this strongly increases the computational cost, because the classifier or similarity function has to be evaluated over a large set of candidate subwindows.In this paper, we propose a simple yet powerful branch and bound scheme that allows efficient maximization of a large class of quality functions over all possible subimages. It converges to a globally optimal solution typically in linear or even sublinear time, in constrast to the quadratic scaling of exhaustive or sliding window search. We show how our method is applicable to different object detection and image retrieval scenarios. The achieved speedup allows the use of classifiers for localization that formerly were considered too slow for this task, such as SVMs with a spatial pyramid kernel or nearest neighbor classifiers based on the χ 2 -distance. We demonstrate state-ofthe-art localization performance of the resulting systems on the UIUC Cars dataset, the PASCAL VOC 2006 dataset and in the PASCAL VOC 2007 competition.

Structured Prediction for Object Detection in Deep Neural Networks

“…can be searched efficiently. In the following description, we draw heavily on Lampert et al [12,13], and adapt their formalism to neural networks.…”

Tactile-Object Recognition From Appearance Information

“…We follow many others in the vision community (e.g. [43]- [45]) by applying only the descriptor portion of the SIFT algorithm to characterize image patches. This practice seems particularly appropriate since the tactile images already represent patches of the object surface.…”

Saliency Density and Edge Response Based Salient Object Detection

“…Recently, ESS [5] based on branch-and-bound search scheme is used to replace the exhaustive search for speeding up the salient object detection process. Though ESS is a general search tool, the formulated objective function for optimization is the key to affect the search accuracy.…”

“…So the distance between the searching window and its inner window is set to one eighth of the image maximum side in our approach. The upper boundF(W) has the desired properties [5]…”

“…Liu et al [3] used exhaustive search to find the optimal window from the thresholded saliency map. To avoid the brute force search, Valenti et al [4] applied efficient sub-window search (ESS) [5] to speed up the search process. Nevertheless, the results of the both methods [3], [4] highly rely on the threshold choice.…”