论文:Improving Object DetectionWith One Line of Code
论文链接:https://arxiv.org/abs/1704.04503
这篇ICCV2017的文章,是NMS算法的改进,从论文题目可以看出,改进仅仅花了一行代码!首先NMS(Non maximum suppression)是在object detection算法中必备的后处理步骤,目的是用来去除重复框,也就是降低误检(false positives)。NMS算法的大致过程可以看原文这段话:First, it sorts all detection boxes on the basis of their scores. The detection box M with the maximum score is selected and all other detection boxes with a significant overlap (using a pre-defined threshold) with M are suppressed. This process is recursively applied on the remaining boxes.
那么传统的NMS算法存在什么问题呢?可以看Figure1。在Fiugre1中,检测算法本来应该输出两个框,但是传统的NMS算法可能会把score较低的绿框过滤掉(如果绿框和红框的IOU大于设定的阈值就会被过滤掉),导致只检测出一个object(一个马),显然这样object的recall就比较低了。