Next Bit Prediction: A Unified Lossless and Lossy Point Cloud Geometry Compression Framework.

Area of Science:

• Computer Vision
• 3D Geometry Processing
• Data Compression

Background:

• 3D point clouds are essential for various applications.
• Efficiently compressing and reconstructing point cloud geometry remains a challenge.
• Existing methods often optimize for either lossless compression or lossy reconstruction, not both simultaneously.

Purpose of the Study:

• To introduce Next Bit Prediction (NBP), a novel framework for unified lossless compression and lossy reconstruction of 3D point clouds.
• To leverage next-bit probability estimation for enhanced geometric symbol prediction.
• To achieve state-of-the-art performance in both compression bitrate and reconstruction quality.

Main Methods:

• Developed a unified framework (NBP) based on next-bit probability estimation.
• Introduced a multi-stage Occupancy Probability Estimation (OPE) for significant bits and high-density regions.
• Proposed a Disentangled Probability Estimation (DPE) module for less significant bits, handling density and residuals.

Main Results:

• NBP achieves superior coding efficiency for lossless compression.
• The framework enables high-fidelity lossy reconstruction through probability-driven refinement.
• Demonstrated state-of-the-art quantitative and qualitative results across extensive experiments.
• NBP exhibits low complexity and progressive coding capabilities.

Conclusions:

• NBP provides a unified and efficient solution for 3D point cloud geometry processing.
• Accurate probability estimation is key to simultaneously optimizing lossless compression and lossy reconstruction.
• The proposed OPE and DPE modules effectively handle varying point densities and bit significance.