Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Efficient Implicit Neural Compression of Point Clouds via Learnable Activation in Latent Space

Yichi Zhang, Qianqian Yang

TL;DR

This work tackles the challenge of efficiently compressing LiDAR-derived point clouds by reframing PCC as neural parameter compression. It introduces PICO, a two-stage framework that separately optimizes geometry and attribute INRs, and LeAFNet, a Kolmogorov-Arnold Network–inspired backbone with learnable activations to improve implicit function fitting while remaining parameter-efficient. The method integrates adaptive sampling, a dynamic occupancy threshold, voxel-wise color regression with nearest-neighbor mapping, and a quantization-then-entropy coding pipeline (DeepCABAC) to produce compact bitstreams. Empirical results on the 8iVFB dataset show substantial RD gains over MPEG G-PCC and V-PCC for both geometry and joint geometry-attribute compression, validating the approach and its potential to influence future intelligent PCC standards.

Abstract

Implicit Neural Representations (INRs), also known as neural fields, have emerged as a powerful paradigm in deep learning, parameterizing continuous spatial fields using coordinate-based neural networks. In this paper, we propose \textbf{PICO}, an INR-based framework for static point cloud compression. Unlike prevailing encoder-decoder paradigms, we decompose the point cloud compression task into two separate stages: geometry compression and attribute compression, each with distinct INR optimization objectives. Inspired by Kolmogorov-Arnold Networks (KANs), we introduce a novel network architecture, \textbf{LeAFNet}, which leverages learnable activation functions in the latent space to better approximate the target signal's implicit function. By reformulating point cloud compression as neural parameter compression, we further improve compression efficiency through quantization and entropy coding. Experimental results demonstrate that \textbf{LeAFNet} outperforms conventional MLPs in INR-based point cloud compression. Furthermore, \textbf{PICO} achieves superior geometry compression performance compared to the current MPEG point cloud compression standard, yielding an average improvement of $4.92$ dB in D1 PSNR. In joint geometry and attribute compression, our approach exhibits highly competitive results, with an average PCQM gain of $2.7 \times 10^{-3}$.

Efficient Implicit Neural Compression of Point Clouds via Learnable Activation in Latent Space

TL;DR

This work tackles the challenge of efficiently compressing LiDAR-derived point clouds by reframing PCC as neural parameter compression. It introduces PICO, a two-stage framework that separately optimizes geometry and attribute INRs, and LeAFNet, a Kolmogorov-Arnold Network–inspired backbone with learnable activations to improve implicit function fitting while remaining parameter-efficient. The method integrates adaptive sampling, a dynamic occupancy threshold, voxel-wise color regression with nearest-neighbor mapping, and a quantization-then-entropy coding pipeline (DeepCABAC) to produce compact bitstreams. Empirical results on the 8iVFB dataset show substantial RD gains over MPEG G-PCC and V-PCC for both geometry and joint geometry-attribute compression, validating the approach and its potential to influence future intelligent PCC standards.

Abstract

Implicit Neural Representations (INRs), also known as neural fields, have emerged as a powerful paradigm in deep learning, parameterizing continuous spatial fields using coordinate-based neural networks. In this paper, we propose \textbf{PICO}, an INR-based framework for static point cloud compression. Unlike prevailing encoder-decoder paradigms, we decompose the point cloud compression task into two separate stages: geometry compression and attribute compression, each with distinct INR optimization objectives. Inspired by Kolmogorov-Arnold Networks (KANs), we introduce a novel network architecture, \textbf{LeAFNet}, which leverages learnable activation functions in the latent space to better approximate the target signal's implicit function. By reformulating point cloud compression as neural parameter compression, we further improve compression efficiency through quantization and entropy coding. Experimental results demonstrate that \textbf{LeAFNet} outperforms conventional MLPs in INR-based point cloud compression. Furthermore, \textbf{PICO} achieves superior geometry compression performance compared to the current MPEG point cloud compression standard, yielding an average improvement of dB in D1 PSNR. In joint geometry and attribute compression, our approach exhibits highly competitive results, with an average PCQM gain of .

Paper Structure

This paper contains 33 sections, 17 equations, 3 figures, 1 table, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Work
  3. Point Cloud Compression
  4. Implicit Neural Representation
  5. Learnable Activation Functions
  6. Methodology
  7. Preliminaries
  8. Geometry Compression
  9. Problem Modeling
  10. Sampling Strategy
  11. Dynamic Threshold
  12. Attribute Compression
  13. Learnable Activation Function Network
  14. Positional Encoding
  15. Learnable Activation Layer
  16. ...and 18 more sections

Figures (3)

  • Figure 1: Model Architecture of LeAFNet.
  • Figure 2: Rate-distortion curves of PICO, PICO (MLP), G-PCC (octree), G-PCC (trisoup) and V-PCC on Static PCC.
  • Figure 3: (a) PCQM RD curve of LeAFNet with different parameter sizes. (b) Unimodal property of $\mathcal{D}$. (c) PSNR RD curve of LeAFNet with different positional encoding lengths $L$. (d) Impact of different sampling strategies on compression performance.