Context extraction method, system, device and medium for point cloud geometry compression
By removing the upsampling operation in point cloud geometric compression and adopting the channel autoregressive method, the problems of limited receptive field and computational redundancy in sparse convolution are solved, achieving more efficient point cloud compression and improved coding performance.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- UNIV OF SCI & TECH OF CHINA
- Filing Date
- 2025-01-17
- Publication Date
- 2026-07-17
AI Technical Summary
In existing point cloud geometric compression technologies, the limited receptive field of sparse convolution and the redundancy of multi-stage autoregressive computation lead to low compression efficiency, affecting the storage and transmission efficiency of point cloud data.
By removing the upsampling operation, the occupancy symbols are stored in the feature channels of the sparse vector, and a multi-stage channel autoregression method is used to extract contextual information, keeping the receptive field of the sparse convolution unchanged and reducing computational redundancy.
It improves the accuracy and efficiency of point cloud geometry compression, reduces computational complexity, and enhances coding performance and compression ratio.
Smart Images
Figure CN122415759A_ABST