基于图注意力网络的自监督3D场景流估计方法
By combining graph attention networks and self-supervised loss functions, the problem of difficult annotation of point cloud datasets is solved, achieving more efficient 3D scene flow estimation and improving detection and segmentation performance.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- HARBIN ENG UNIV
- Filing Date
- 2022-11-22
- Publication Date
- 2026-07-17
AI Technical Summary
Existing technologies struggle to effectively estimate 3D scene flow, especially when point cloud datasets are difficult to annotate and point cloud feature information is hard to capture, resulting in insufficient 3D motion detection and segmentation performance.
A graph attention network (AGMNet) is used in conjunction with a self-supervised loss function. Local feature information is extracted by constructing a topological structure, and self-supervised methods are used for training, including graph attention convolution and attention pooling structures. Combined with recurrent cost layers and cross-scale attention structures, a self-supervised loss function is constructed to improve network performance.
It improves the estimation accuracy and network performance of 3D scene flow, reduces the dependence on labeled data, and enhances the efficiency of point cloud feature matching and scene flow estimation.
Smart Images
Figure CN115797557B_ABST