Weakly Supervised Semantic Roadside Object Segmentation Using Digital Maps

Abstract

Publicly available digital maps may offer semantic information regarding objects in street view images. In this paper, we propose an approach to exploit such information to automatically create object detection datasets on which state-of-the-art object detection methods can be trained. To accomplish this, we use two detailed maps of the Netherlands containing the location of a large number of street objects. We link the object information to street view images to use them as image-wide labels. Our results show that even though there are many sources for noise in the labels, we can create useful data with this approach.