NEW-GPCC: A Configurable Neural Empowered Wrapper for Geometry-Based Point Cloud Compression

Geometry-based Point Cloud Compression (G-PCC) is an MPEG-standardized point cloud codec that offers cross-platform compatibility and robustness validated within large expert communities. Despite its generalizability and reasonable balance between performance and complexity, the compression efficiency of G-PCC lags behind recent deep learning-based approaches. To bridge this gap, this paper introduces NEW-GPCC, a configurable Neural Empowered Wrapper designed to enhance G-PCC while fully preserving its standard compliance and low complexity. The wrapper comprises a multi-level Compact Representation Network (CRNet) for pre-processing, a multi-level Learned LookUp Table (LLUT), and a Quality Enhancement Network (QENet) for post-processing. The multi-level configuration design enables flexible tradeoffs between complexity and efficiency. By jointly optimizing the wrapper with the proposed differentiable G-PCC surrogate, the wrapper learns to adapt to the intrinsic rate–distortion characteristics of G-PCC, thereby achieving optimal end-to-end coding efficiency. Experimental results show an average 81.58% BD-rate reduction over G-PCC without altering the bitstream syntax or decoder, highlighting a practical pathway to advance standardized codecs with neural technologies while maintaining interoperability.