Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

TSPE-GS: Probabilistic Depth Extraction for Semi-Transparent Surface Reconstruction via 3D Gaussian Splatting

Zhiyuan Xu, Nan Min, Yuhang Guo, Tong Wei

TL;DR

TSPE-GS addresses the failure of single-depth assumptions in 3D Gaussian Splatting when reconstructing semi-transparent scenes. By modeling per-pixel depth as a multi-modal distribution over sampled transmittances and extracting multiple surface depths via peak detection, it enables simultaneous outer and inner surface reconstruction. The method integrates these multi-layer depths through progressive TSDF fusion, preventing interference between layers. Experiments on semi-transparent and opaque datasets show significant gains in semi-transparent geometry without sacrificing opaque performance, and the approach remains a plug-and-play enhancement for Gaussian-based pipelines.

Abstract

3D Gaussian Splatting offers a strong speed-quality trade-off but struggles to reconstruct semi-transparent surfaces because most methods assume a single depth per pixel, which fails when multiple surfaces are visible. We propose TSPE-GS (Transparent Surface Probabilistic Extraction for Gaussian Splatting), which uniformly samples transmittance to model a pixel-wise multi-modal distribution of opacity and depth, replacing the prior single-peak assumption and resolving cross-surface depth ambiguity. By progressively fusing truncated signed distance functions, TSPE-GS reconstructs external and internal surfaces separately within a unified framework. The method generalizes to other Gaussian-based reconstruction pipelines without extra training overhead. Extensive experiments on public and self-collected semi-transparent and opaque datasets show TSPE-GS significantly improves semi-transparent geometry reconstruction while maintaining performance on opaque scenes.

TSPE-GS: Probabilistic Depth Extraction for Semi-Transparent Surface Reconstruction via 3D Gaussian Splatting

TL;DR

TSPE-GS addresses the failure of single-depth assumptions in 3D Gaussian Splatting when reconstructing semi-transparent scenes. By modeling per-pixel depth as a multi-modal distribution over sampled transmittances and extracting multiple surface depths via peak detection, it enables simultaneous outer and inner surface reconstruction. The method integrates these multi-layer depths through progressive TSDF fusion, preventing interference between layers. Experiments on semi-transparent and opaque datasets show significant gains in semi-transparent geometry without sacrificing opaque performance, and the approach remains a plug-and-play enhancement for Gaussian-based pipelines.

Abstract

3D Gaussian Splatting offers a strong speed-quality trade-off but struggles to reconstruct semi-transparent surfaces because most methods assume a single depth per pixel, which fails when multiple surfaces are visible. We propose TSPE-GS (Transparent Surface Probabilistic Extraction for Gaussian Splatting), which uniformly samples transmittance to model a pixel-wise multi-modal distribution of opacity and depth, replacing the prior single-peak assumption and resolving cross-surface depth ambiguity. By progressively fusing truncated signed distance functions, TSPE-GS reconstructs external and internal surfaces separately within a unified framework. The method generalizes to other Gaussian-based reconstruction pipelines without extra training overhead. Extensive experiments on public and self-collected semi-transparent and opaque datasets show TSPE-GS significantly improves semi-transparent geometry reconstruction while maintaining performance on opaque scenes.

Paper Structure

This paper contains 34 sections, 17 equations, 13 figures, 4 tables, 2 algorithms.

Table of Contents

  1. Introduction
  2. Related Work
  3. Preliminaries
  4. The Proposed Method
  5. Probabilistic depth modeling
  6. Unified surface depth hypothesis
  7. Obtaining the Probability Distribution
  8. Calculating the Peak Locations
  9. Fuse Peak Depths
  10. Experiments
  11. Result Comparison
  12. Visual Comparison
  13. Hyperparameter Analysis
  14. Conclusion
  15. Appendix A: Experimental Verification of the Unimodality of Depth Distributions
  16. ...and 19 more sections

Figures (13)

  • Figure 1: Visual comparison of reconstruction on a semi-transparent object in $\alpha$Surf dataset with and without TSPE.
  • Figure 2: Depth PDF and CDF for opaque surfaces.
  • Figure 3: Depth PDF and CDF for semi-transparent surfaces.
  • Figure 4: Workflow of the TSPE-GS pipeline for multi-layer depth estimation and reconstruction.
  • Figure 5: We visually compare our method with other Gaussian-based geometry reconstruction pipelines to demonstrate the effectiveness.
  • ...and 8 more figures