Learning to Produce Semi-dense Correspondences for Visual Localization | Read Paper on Bytez