Efficient Large-Scale Point Cloud Geometry Compression.

Area of Science:

• Computer Vision
• Data Compression
• 3D Graphics

Background:

• Large-scale point clouds demand significant bandwidth and memory.
• Existing point cloud geometry compression methods face challenges in performance and complexity.

Purpose of the Study:

• To propose an efficient algorithm for large-scale point cloud geometry compression.
• To address suboptimal compression performance and complex encoding-decoding processes.

Main Methods:

• Designed a cross-attention module in the encoder to incorporate positional information and enhance features.
• Introduced an efficient generation module in the decoder to improve quality without increasing time.

Main Results:

• Achieved an average bitrate reduction of -46.64% compared to G-PCC v23.
• Demonstrated the fastest decoding time among state-of-the-art methods.
• Maintained a minimal network model size of 2.8 M.

Conclusions:

• The proposed algorithm offers superior compression performance for large-scale point clouds.
• The method provides significant bitrate savings and faster decoding, making it practical for real-world applications.