Weakly Supervised Semantic Segmentation Based on Co-segmentation

Training a Fully Convolutional Network (FCN) for semantic segmentation requires a large number of pixel-level masks, which involves a large amount of human labour and time for annotation. In contrast, image-level labels are much easier to obtain. In this work, we propose a novel method for weakly supervised semantic segmentation with only image-level labels. The method relies on a large scale co-segmentation framework that can produce object masks for a group of images containing objects belonging to the same semantic class. We first retrieve images from search engines, e.g. Flickr and Google, using semantic class names as queries, e.g. class names in PASCAL VOC 2012. We then use high quality masks produced by co-segmentation on the retrieved images as well as the target dataset images with image level labels to train segmentation networks. We obtain IoU 56.9 on test set of PASCAL VOC 2012, which reaches state of the art performance.