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

2025-11-13

AI Summary

3D Gaussian Splatting suffers from geometric ambiguity in translucent object reconstruction due to its inherent single-depth assumption, leading to inter-surface confusion and depth uncertainty. To address this, we propose a probabilistic depth extraction framework that models pixel-wise multimodal opacity and estimates depth distributions to decouple multiple surface layers. We further integrate uniform transmittance sampling with a truncated signed distance function (TSDF) to jointly optimize geometry and appearance within the 3D Gaussian Splatting pipeline. Our method requires no auxiliary neural network training and preserves the original real-time rendering efficiency. Evaluated on both translucent and opaque datasets, it significantly improves geometric fidelity and visual quality. Notably, it achieves, for the first time in explicit radiance fields, fine-grained separation and reconstruction of interior and exterior surfaces of transparent objects.

Technology Category

• Computer Vision: 3D Computer Vision
• Search and Optimization: Sampling/Simulation-based Search
• Machine Learning: Transparent, Interpretable, Explainable ML

Application Category

• Security and Privacy: Data transparency and provenance
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG

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.

Problem

• Reconstructing semi-transparent surfaces using 3D Gaussian Splatting
• Resolving depth ambiguity when multiple surfaces are visible
• Modeling pixel-wise multi-modal opacity and depth distributions

Innovation

• Uniformly samples transmittance for multi-modal depth distribution
• Progressively fuses truncated signed distance functions for surfaces
• Generalizes to Gaussian pipelines without extra training overhead