Image Segmentation has attracted much attention in deep learning, particularly in case of medical field in which it can help in building computer-aided diagnosis systems. Of all the methods that have been designed for image segmentation or salient object detection, UNet and architectures based on it have attracted much attention and have been highly accurate in segmentation tasks. There have been numerous architectures that are based on UNet original architecture, each one employing its own unique adaptation. In this paper, we investigate the usefulness of three UNet-based architectures: UNet3+, Attention UNet and \(U^2\) Net in segmentation of Skin lesion images.