Concetti Chiave
Leveraging geometric priors enhances human point cloud compression performance.
Sintesi
The study focuses on leveraging geometric priors for compressing high-resolution human point clouds efficiently. It introduces a novel framework that combines a geometric prior with structure variations to improve coding performance. The methodology involves two stages: geometric prior representation and feature residual extraction and compression. Extensive experiments demonstrate the superiority of the proposed approach over traditional and learning-based methods, showcasing significant bitrate savings and PSNR gains across various datasets.
- Introduction to the demand for human point clouds in extended reality.
- Challenges in compressing high-resolution human point clouds.
- Overview of traditional PCC methods by MPEG and deep learning-based techniques.
- Proposal of a novel framework leveraging geometric priors for compression.
- Detailed explanation of the methodology involving geometric prior representation and feature residual extraction.
- Training details, including datasets used and performance evaluation metrics.
- Comparison of results against traditional (G-PCC, V-PCC) and learning-based (PCGC, PCGCv2) methods.
- Performance comparisons showing significant improvements in bitrate savings and PSNR gains.
Statistiche
A high-resolution human point cloud comprises 765,000 points with 30-bit coordinates and 24-bit color information.
The proposed framework achieves an average bitrate saving of 92.34% compared to G-PCC (octree).
Our approach outperforms PCGCv2 by approximately 1.26 dB on various datasets.
Citazioni
"Our main contributions are summarized as follows: We propose a novel geometric prior based point cloud geometry compression framework."
"Extensive experimental results show that our approach significantly improves the compression performance without deteriorating the quality."