Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Faithful Contouring: Near-Lossless 3D Voxel Representation Free from Iso-surface

Yihao Luo, Xianglong He, Chuanyu Pan, Yiwen Chen, Jiaqi Wu, Yangguang Li, Wanli Ouyang, Yuanming Hu, Guang Yang, ChoonHwai Yap

TL;DR

Faithful Contouring introduces a near-lossless, distance-field-free voxel representation that directly encodes arbitrary meshes into sparse contour tokens (FCTs) and supports $2048^3$ resolution without iso-surface extraction. The method replaces traditional SDF/occupancy pipelines with a local, GPU-friendly remeshing approach, enabling open surfaces, non-manifold geometries, and internal cavities to be preserved. A dual-mode VAE validates the representation for both direct token-based reconstruction and point-to-FCT reconstruction, achieving state-of-the-art fidelity and efficiency on challenging datasets. The work enables high-resolution 3D learning and editing tasks with faithful geometry, sharp features, and flexible manipulation, marking a significant shift away from iso-surface-centric workflows. Overall, Faithful Contouring delivers a scalable, high-fidelity alternative for 3D reconstruction and generation that preserves fine details while supporting downstream editing and texturing.

Abstract

Accurate and efficient voxelized representations of 3D meshes are the foundation of 3D reconstruction and generation. However, existing representations based on iso-surface heavily rely on water-tightening or rendering optimization, which inevitably compromise geometric fidelity. We propose Faithful Contouring, a sparse voxelized representation that supports 2048+ resolutions for arbitrary meshes, requiring neither converting meshes to field functions nor extracting the isosurface during remeshing. It achieves near-lossless fidelity by preserving sharpness and internal structures, even for challenging cases with complex geometry and topology. The proposed method also shows flexibility for texturing, manipulation, and editing. Beyond representation, we design a dual-mode autoencoder for Faithful Contouring, enabling scalable and detail-preserving shape reconstruction. Extensive experiments show that Faithful Contouring surpasses existing methods in accuracy and efficiency for both representation and reconstruction. For direct representation, it achieves distance errors at the $10^{-5}$ level; for mesh reconstruction, it yields a 93\% reduction in Chamfer Distance and a 35\% improvement in F-score over strong baselines, confirming superior fidelity as a representation for 3D learning tasks.

Faithful Contouring: Near-Lossless 3D Voxel Representation Free from Iso-surface

TL;DR

Faithful Contouring introduces a near-lossless, distance-field-free voxel representation that directly encodes arbitrary meshes into sparse contour tokens (FCTs) and supports resolution without iso-surface extraction. The method replaces traditional SDF/occupancy pipelines with a local, GPU-friendly remeshing approach, enabling open surfaces, non-manifold geometries, and internal cavities to be preserved. A dual-mode VAE validates the representation for both direct token-based reconstruction and point-to-FCT reconstruction, achieving state-of-the-art fidelity and efficiency on challenging datasets. The work enables high-resolution 3D learning and editing tasks with faithful geometry, sharp features, and flexible manipulation, marking a significant shift away from iso-surface-centric workflows. Overall, Faithful Contouring delivers a scalable, high-fidelity alternative for 3D reconstruction and generation that preserves fine details while supporting downstream editing and texturing.

Abstract

Accurate and efficient voxelized representations of 3D meshes are the foundation of 3D reconstruction and generation. However, existing representations based on iso-surface heavily rely on water-tightening or rendering optimization, which inevitably compromise geometric fidelity. We propose Faithful Contouring, a sparse voxelized representation that supports 2048+ resolutions for arbitrary meshes, requiring neither converting meshes to field functions nor extracting the isosurface during remeshing. It achieves near-lossless fidelity by preserving sharpness and internal structures, even for challenging cases with complex geometry and topology. The proposed method also shows flexibility for texturing, manipulation, and editing. Beyond representation, we design a dual-mode autoencoder for Faithful Contouring, enabling scalable and detail-preserving shape reconstruction. Extensive experiments show that Faithful Contouring surpasses existing methods in accuracy and efficiency for both representation and reconstruction. For direct representation, it achieves distance errors at the level; for mesh reconstruction, it yields a 93\% reduction in Chamfer Distance and a 35\% improvement in F-score over strong baselines, confirming superior fidelity as a representation for 3D learning tasks.

Paper Structure

This paper contains 30 sections, 13 equations, 7 figures, 2 tables, 2 algorithms.

Table of Contents

  1. Introduction
  2. Related Work
  3. Method: Faithful Contouring
  4. FCT Representation and Remeshing
  5. Encoder (i): Active Voxel Detection.
  6. Encoder (ii): Intersection Centroids.
  7. Encoder (iii): Anchor Fitting.
  8. Matrix form and closed-form solvers.
  9. Encoder (iv): Semi-axis Intersections.
  10. Faithful Contour Tokens (FCT).
  11. Decoder: Gathering & Remeshing.
  12. Editing of FCT
  13. Filtering
  14. Texture
  15. Manipulation
  16. ...and 15 more sections

Figures (7)

  • Figure 1: Faithful Contouring: A Near-Lossless Voxelized 3D Representation keeps fine-grained geometric details while maintaining internal structure. This representation encodes an arbitrary mesh into voxelized tokens, supporting 2048+ resolution with neither iso-surface extraction from the converted SDFs nor differentiable rendering optimization. Please zoom in to view the detailed geometry from the remeshing results.
  • Figure 2: Comparison of representing pipelines.Traditional UDF $\rightarrow$ water-tightening $\rightarrow$ SDF $\rightarrow$ iso-surface pipelines, relying on Marching Cubes and its variants, introduce artifacts at each lossy step, including artificial surface thickening, loss of internal structures, and jagged iso-surface extraction. In contrast, Faithful Contouring directly obtains voxelized features, including fitted anchors and connections, from raw meshes with a highly accurate remeshing algorithm.
  • Figure 3: Faithful Contour pipeline.Encoder voxelizes the input mesh, then computes centroids, anchors, and semi-axis intersections, and stores them in the Faithful Contour Token (FCT) on $K$ active voxels. Decoder gathers anchors, resolves orientations, and remeshes the tokens into high-fidelity surfaces.
  • Figure 4: Sharpness from low–resolution reconstruction with Faithful Contouring. Ground-truth surface (GT) compared with reconstructions at voxel resolutions of $8^3$, $16^3$, $32^3$, and $64^3$. Despite coarse discretization, our method preserves overall shape and captures sharp geometric features, with error visualized in red.
  • Figure 5: Demonstration of FCT Editing.Assembly of two geometric components, subsequent Manipulation (transformation/posing) of the combined object, and the texture can be recovered by voxel-wise RGB features attached on FCT.
  • ...and 2 more figures