DualSplat: Robust 3D Gaussian Splatting via Pseudo-Mask Bootstrapping from Reconstruction Failures
For 3D reconstruction from multi-view images with transient objects, DualSplat provides a robust solution that breaks the circular dependency between transient detection and static scene reconstruction.
DualSplat addresses the performance degradation of 3D Gaussian Splatting in scenes with transient objects by converting first-pass reconstruction failures into pseudo-masks for a second-pass clean optimization. It achieves state-of-the-art results on RobustNeRF and NeRF On-the-go, especially in transient-heavy scenes.
While 3D Gaussian Splatting (3DGS) achieves real-time photorealistic rendering, its performance degrades significantly when training images contain transient objects that violate multi-view consistency. Existing methods face a circular dependency: accurate transient detection requires a well-reconstructed static scene, while clean reconstruction itself depends on reliable transient masks. We address this challenge with DualSplat, a Failure-to-Prior framework that converts first-pass reconstruction failures into explicit priors for a second reconstruction stage. We observe that transients, which appear in only a subset of views, often manifest as incomplete fragments during conservative initial training. We exploit these failures to construct object-level pseudo-masks by combining photometric residuals, feature mismatches, and SAM2 instance boundaries. These pseudo-masks then guide a clean second-pass 3DGS optimization, while a lightweight MLP refines them online by gradually shifting from prior supervision to self-consistency. Experiments on RobustNeRF and NeRF On-the-go show that DualSplat outperforms existing baselines, demonstrating particularly clear advantages in transient-heavy scenes and transient regions.