no code implementations • ICCV 2023 • Xiaoyong Lu, Yaping Yan, Tong Wei, Songlin Du
Current feature matching methods focus on point-level matching, pursuing better representation learning of individual features, but lacking further understanding of the scene.
no code implementations • 2 Mar 2023 • Xiaoyong Lu, Yaping Yan, Bin Kang, Songlin Du
Heavy computation is a bottleneck limiting deep-learningbased feature matching algorithms to be applied in many realtime applications.
no code implementations • 20 Oct 2021 • Haiyang Liu, Dingli Luo, Songlin Du, Takeshi Ikenaga
To solve these problems, this paper proposes (i) a new loss organization method which uses self-supervised heatmaps to reduce prediction contradictions and spatial-sequential attention to enhance networks' features extraction; (ii) a new combination of predictions composed by heatmaps, Part Affinity Fields (PAFs) and our block-inside offsets to fix pixel-level joints positions and further demonstrates the effectiveness of proposed loss function.