Self-supervised Learning of Dense Visual Descriptors