Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

SurfelSoup: Learned Point Cloud Geometry Compression With a Probablistic SurfelTree Representation

Tingyu Fan, Ran Gong, Yueyu Hu, Yao Wang

TL;DR

SurfelSoup is presented, an end-to-end learned surface-based framework for point cloud geometry compression, with surface-structured primitives for representation, which proposes a probabilistic surface representation, pSurfel, which models local point occupancies using a bounded generalized Gaussian distribution.

Abstract

This paper presents SurfelSoup, an end-to-end learned surface-based framework for point cloud geometry compression, with surface-structured primitives for representation. It proposes a probabilistic surface representation, pSurfel, which models local point occupancies using a bounded generalized Gaussian distribution. In addition, the pSurfels are organized into an octree-like hierarchy, pSurfelTree, with a Tree Decision module that adaptively terminates the tree subdivision for rate-distortion optimal Surfel granularity selection. This formulation avoids redundant point-wise compression in smooth regions and produces compact yet smooth surface reconstructions. Experimental results under the MPEG common test condition show consistent gain on geometry compression over voxel-based baselines and MPEG standard G-PCC-GesTM-TriSoup, while providing visually superior reconstructions with smooth and coherent surface structures.

SurfelSoup: Learned Point Cloud Geometry Compression With a Probablistic SurfelTree Representation

TL;DR

SurfelSoup is presented, an end-to-end learned surface-based framework for point cloud geometry compression, with surface-structured primitives for representation, which proposes a probabilistic surface representation, pSurfel, which models local point occupancies using a bounded generalized Gaussian distribution.

Abstract

This paper presents SurfelSoup, an end-to-end learned surface-based framework for point cloud geometry compression, with surface-structured primitives for representation. It proposes a probabilistic surface representation, pSurfel, which models local point occupancies using a bounded generalized Gaussian distribution. In addition, the pSurfels are organized into an octree-like hierarchy, pSurfelTree, with a Tree Decision module that adaptively terminates the tree subdivision for rate-distortion optimal Surfel granularity selection. This formulation avoids redundant point-wise compression in smooth regions and produces compact yet smooth surface reconstructions. Experimental results under the MPEG common test condition show consistent gain on geometry compression over voxel-based baselines and MPEG standard G-PCC-GesTM-TriSoup, while providing visually superior reconstructions with smooth and coherent surface structures.
Paper Structure (67 sections, 10 equations, 17 figures, 4 tables, 2 algorithms)

This paper contains 67 sections, 10 equations, 17 figures, 4 tables, 2 algorithms.

Table of Contents

  1. Introduction
  2. Related Work
  3. Rule-based Point Cloud Compression
  4. Learning-based Point Cloud Compression
  5. 3D Representations
  6. Methodology
  7. Overview
  8. Encoder & Entropy Coding
  9. pSurfel Representation
  10. Bounding Box.
  11. Voxelization and Distortion.
  12. Binarization.
  13. pSurfelTree Decision
  14. Octree Node Occupancy Compression: P-SOPA
  15. Loss Function Formulated as Expectations
  16. ...and 52 more sections

Figures (17)

  • Figure 1: An illustration of SurfelSoup. Top Left: a toy 2D example for the pSurfelTree structure. Top Right: Example reconstruction comparison of a 3D Husky. Bottom: Example of pSurfelTree's multi-granularity reconstruction.
  • Figure 2: The overall architecture of SurfelSoup. Encoder and Entropy Model correspond to Sec. \ref{['chap:encoder']}. Surfel Reconstruction, Voxelization and Binarization correspond to Sec. \ref{['chap:surfel_recon']}. Decision corresponds to Sec. \ref{['chap:decision']}. Octree Compression corresponds to Sec. \ref{['chap:SOPA']}.
  • Figure 3: A 2D example to explain P-SOPA. The octants labeled 1 and 2 have been coded, with gray intensities indicating the existence probabilities (the darker the higher probability). Octants labeled 3 are being coded, and octants labeled 4 are yet to be coded and unseen. In the middle figure, some octants labeled 1 and 2 were randomly masked out (set as white). Real P-SOPA is done on 3D.
  • Figure 4: The comparison between SurfelSoup (ours) and baselines on Owlii.
  • Figure 5: The comparison between SurfelSoup (ours) and baselines on RWTT dataset.
  • ...and 12 more figures