Direct Object-Level Reconstruction via Probabilistic Gaussian Splatting

2026-03-15

AI Summary

This work addresses the excessive computational and storage overhead in existing Gaussian splatting methods caused by redundant background elements in full-scene reconstruction. To this end, we propose an efficient single-object 3D reconstruction approach that integrates foreground-background probability cues into Gaussian primitives. By replacing conventional binary masks with continuous probability masks and incorporating a two-stage filtering strategy alongside a probability-supervised self-correction mechanism, our method effectively mitigates boundary blurriness and suppresses background interference. Leveraging probability masks generated by YOLO and SAM to guide Gaussian attribute learning, combined with dynamic pruning and rendering-mask feedback optimization, our approach achieves reconstruction quality comparable to standard 3DGS using only approximately one-tenth the number of Gaussians on the MIP-360, T&T, and NVOS datasets, while demonstrating strong robustness to mask inaccuracies.

Technology Category

• Computer Vision: 3D Computer Vision
• Machine Learning: Feature Construction/Reformulation
• Search and Optimization: Mixed Discrete/Continuous Search

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Economics, Online Markets and Human Computation: LLM based quality controls for crowd work
• Security and Privacy: Data transparency and provenance

Abstract

Object-level 3D reconstruction play important roles across domains such as cultural heritage digitization, industrial manufacturing, and virtual reality. However, existing Gaussian Splatting-based approaches generally rely on full-scene reconstruction, in which substantial redundant background information is introduced, leading to increased computational and storage overhead. To address this limitation, we propose an efficient single-object 3D reconstruction method based on 2D Gaussian Splatting. By directly integrating foreground-background probability cues into Gaussian primitives and dynamically pruning low-probability Gaussians during training, the proposed method fundamentally focuses on an object of interest and improves the memory and computational efficiency. Our pipeline leverages probability masks generated by YOLO and SAM to supervise probabilistic Gaussian attributes, replacing binary masks with continuous probability values to mitigate boundary ambiguity. Additionally, we propose a dual-stage filtering strategy for training's startup to suppress background Gaussians. And, during training, rendered probability masks are conversely employed to refine supervision and enhance boundary consistency across views. Experiments conducted on the MIP-360, T&T, and NVOS datasets demonstrate that our method exhibits strong self-correction capability in the presence of mask errors and achieves reconstruction quality comparable to standard 3DGS approaches, while requiring only approximately 1/10 of their Gaussian amount. These results validate the efficiency and robustness of our method for single-object reconstruction and highlight its potential for applications requiring both high fidelity and computational efficiency.

Problem

• object-level reconstruction
• Gaussian Splatting
• background redundancy
• computational efficiency
• 3D reconstruction

Innovation

• Probabilistic Gaussian Splatting
• Object-level Reconstruction
• Probability Mask
• Dynamic Pruning
• Boundary Consistency