Weakly Supervised Learning of Instance Segmentation with Inter-pixel Relations

This paper presents a novel approach for learning instance segmentation withimage-level class labels as supervision. Our approach generates pseudo instancesegmentation labels of training images, which are used to train a fullysupervised model. For generating the pseudo labels, we first identify confidentseed areas of object classes from attention maps of an image classificationmodel, and propagate them to discover the entire instance areas with accurateboundaries. To this end, we propose IRNet, which estimates rough areas ofindividual instances and detects boundaries between different object classes.It thus enables to assign instance labels to the seeds and to propagate themwithin the boundaries so that the entire areas of instances can be estimatedaccurately. Furthermore, IRNet is trained with inter-pixel relations on theattention maps, thus no extra supervision is required. Our method with IRNetachieves an outstanding performance on the PASCAL VOC 2012 dataset, surpassingnot only previous state-of-the-art trained with the same level of supervision,but also some of previous models relying on stronger supervision.