The success of deep learning in computer vision is partly attributed to the construction of large-scale annotated datasets such as ImageNet. However, computer vision problems often require substantial amount of human efforts to obtain accurate annotations due to dynamic aspects of class labels, needs for pixel-level labeling, and annotation ambiguities. Hence, collecting high quality large-scale annotated datasets is very time consuming and even unrealistic. This talk discusses a semantic segmentation problem, which can derive benefit from weakly supervised learning. Specifically, I present three weakly supervised semantic segmentation algorithms investigated in POSTECH Computer Vision Laboratory–a semi-supervised few-shot learning method, a transfer learning approach, and a technique based on superpixel pooling network. All algorithms have unique architectures of convolutional neural networks, and achieve the state-of-the-art performance in PASCAL VOC dataset.