This paper presents a weakly supervised sparse learning approach to theproblem of noisily tagged image parsing, or segmenting all the objects within anoisily tagged image and identifying their categories (i.e. tags). Differentfrom the traditional image parsing that takes pixel-level labels as strongsupervisory information, our noisily tagged image parsing is provided withnoisy tags of all the images (i.e. image-level labels), which is a naturalsetting for social image collections (e.g. Flickr). By oversegmenting all theimages into regions, we formulate noisily tagged image parsing as a weaklysupervised sparse learning problem over all the regions, where the initiallabels of each region are inferred from image-level labels. Furthermore, wedevelop an efficient algorithm to solve such weakly supervised sparse learningproblem. The experimental results on two benchmark datasets show theeffectiveness of our approach. More notably, the reported surprising resultsshed some light on answering the question: can image-level labels replacepixel-level labels (hard to access) as supervisory information for imageparsing.
展开▼
