Rectifying Mask via Entropy for Distractor-Free 3DGS in Ambiguous Scenarios

2026-06-28

AI Summary

This work addresses the challenge that existing 3D Gaussian Splatting (3DGS) methods struggle to distinguish transient distractors from static backgrounds in scenes with color or semantic ambiguity, leading to contaminated reconstructions. To resolve this, the authors propose RefineSplat, a novel framework that introduces entropy as a key criterion for identifying ambiguous distractors. By integrating entropy-aware adaptive masking with instance segmentation to detect interference regions, and designing an entropy-guided positional gradient optimization mechanism to regulate Gaussian density distribution, RefineSplat significantly improves novel view synthesis quality across multiple datasets while effectively suppressing distractors. Additionally, the authors construct and publicly release Ambiguous Wild—the first dataset specifically targeting ambiguous interference, comprising 18 challenging real-world scenes—to establish a benchmark for future research in this domain.

Abstract

We present RefineSplat, a systematic framework that effectively constructs transient masks to identify diverse ambiguous distractors. To do this, we qualitatively and quantitatively analyze issues and propose a novel entropy-aware adaptive masking method. Unlike existing approaches that struggle to distinguish transient elements from static scenes due to color or semantic ambiguity, RefineSplat captures ambiguous distractors leveraging entropy and instance masks. Furthermore, we propose a simple yet effective entropy-aware density control to align Gaussians in ambiguous scenarios considering Entropy-aware positional gradients. Additionally, to rigorously validate our method, we first create and release the Ambiguous wild dataset, including 18 scenes where distractors and static scenes are hard to distinguish due to color or semantic resemblances. Experimental results on various datasets demonstrate that RefineSplat shows state-of-the-art performance, showing distractor-free novel view synthesis.

Problem

• ambiguous scenarios
• distractor removal
• 3D Gaussian Splatting
• transient objects
• view synthesis

Innovation

• entropy-aware masking
• 3D Gaussian Splatting
• ambiguous distractors
• novel view synthesis
• adaptive density control