2021/08/19 by Hyunyoung Jung, Eunhyeok Park, Jung, Hyunyoung +3 · 2 citations
Computer Science · Engineering · #Advanced Vision and Imaging #Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences #Image Processing Techniques and Applications #Optical measurement and interference techniques
paper · pdf · doi:10.48550/arxiv.2108.08829
openalex publication_date 2021/08/19 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
Self-supervised monocular depth estimation has been widely studied, owing to\nits practical importance and recent promising improvements. However, most works\nsuffer from limited supervision of photometric consistency, especially in weak\ntexture regions and at object boundaries. To overcome this weakness, we propose\nnovel ideas to improve self-supervised monocular depth estimation by leveraging\ncross-domain information, especially scene semantics. We focus on incorporating\nimplicit semantic knowledge into geometric representation enhancement and\nsuggest two ideas: a metric learning approach that exploits the\nsemantics-guided local geometry to optimize intermediate depth representations\nand a novel feature fusion module that judiciously utilizes cross-modality\nbetween two heterogeneous feature representations. We comprehensively evaluate\nour methods on the KITTI dataset and demonstrate that our method outperforms\nstate-of-the-art methods. The source code is available at\nhttps://github.com/hyBlue/FSRE-Depth.\n