0em Weakly Supervised Object Boundaries-0.7em

State-of-the-art learning based boundary detection methods require extensive training data. Since labelling object boundaries is one of the most expensive types of annotations, there is a need to relax the requirement to carefully annotate images to make both the training more affordable and to extend the amount of training data. In this paper we focus on learning object boundaries in a weakly supervised fashion and show that high quality object boundary detection can be obtained without using any object-specific boundary annotations. We propose several ways of generating object boundary annotations with different levels of supervision, from just using a bounding box oriented object detector to using the boundary detector trained on generic boundaries. For generating weak object boundary annotations we consider different sources, fusing unsupervised image segmentation [5] and object proposal methods [8, 12] with object detectors [6, 9]. We show that bounding box annotations alone suffice to achieve objects boundary estimates with high quality. We present results using a decision forest (SE) [3] and a convnet edge detector (HED) [13]. We report top performance on Pascal object boundary detection [4, 7] with our weak-supervision approaches already surpassing previously reported fully supervised results. Our main contributions are summarized below: We introduce the problem of weakly supervised object-specific boundary detection. We show that good performance can be obtained on BSDS, Pascal VOC12, and SBD boundary estimation using only weak-supervision (leveraging bounding box detection annotations without the need of instance-wise object boundary annotations). We report best known results on PascalVOC12, and SBD datasets. Our weakly supervised results alone improve over the previous fully supervised state-of-the-art. Further information is available at https://goo.gl/kDVZwS.