Weakly- and Semi-Supervised Learning of a Deep Convolutional Network for Semantic Image Segmentation

Deep convolutional neural networks (DCNNs) trained on a large number of images with strong pixel-level annotations have recently sig-nificantly pushed the state-of-art in semantic im-age segmentation. We study the more challeng-ing problem of learning DCNNs for semantic im-age segmentation from either (1) weakly anno-tated training data such as bounding boxes or image-level labels or (2) a combination of few strongly labeled and many weakly labeled im-ages, sourced from one or multiple datasets. We develop methods for semantic image segmenta-tion model training under these weakly super-vised and semi-supervised settings. Extensive experimental evaluation shows that the proposed techniques can learn models delivering state-of-art results on the challenging PASCAL VOC 2012 image segmentation benchmark, while re-quiring significantly less annotation effort. 1