VisDom: Sparse Novel View Synthesis with Visible Domain Constraint

2026-06-18

AI Summary

This work addresses the challenge of novel view synthesis under sparse input views, where geometric ambiguities often lead to floating artifacts and structural inconsistencies. The authors propose VisDom, a learning-free geometric constraint method that introduces, for the first time, minimal multi-view visibility into this task. By defining a visible domain—a spatial subset observed by at least K input views—VisDom refines traditional silhouette carving to filter reconstruction results and strengthen spatial priors. Relying solely on input silhouettes without any learnable parameters, VisDom seamlessly integrates into both NeRF and 3D Gaussian Splatting frameworks, constraining volume sampling and guiding Gaussian optimization. Evaluated on challenging benchmarks such as Omni3D and MipNeRF360 with only four input images, VisDom significantly improves geometric consistency; when applied to GaussianObject, it further enhances performance while reducing training cost by a factor of 22.

Abstract

Sparse novel view synthesis (NVS) remains challenging due to the ambiguity of recovering 3D geometry from few input views. While NeRF- and Gaussian Splatting (GS)-based methods perform well with dense supervision, they often overfit in sparse settings, producing floating artifacts and inconsistent geometry. Silhouette consistency is commonly used as a regularizer, but it remains insufficient, as silhouette-consistent regions can extend beyond the true object geometry. We introduce VisDom, a learning-free geometric constraint that augments classical carving-based visual hull reconstruction by enforcing a minimum multi-view visibility requirement. Specifically, we define a visible domain as the subset of 3D space observed by at least $K$ views and use it as an additional filtering criterion on top of standard silhouette-based reconstruction. This provides a stronger spatial prior in sparse-view settings. We integrate VisDom into both implicit (NeRF) and explicit (GS) pipelines by restricting volumetric sampling and guiding Gaussian placement during optimization. Experiments on three challenging datasets show consistent improvements in sparse-view NVS, enabling high-quality object-centric reconstruction from as few as four input images. Our method is domain-agnostic, requires only silhouettes, and introduces no learned parameters, making it a simple complement to existing approaches. Applying VisDom on top of GaussianObject further improves performance on Omni3D and MipNeRF360, while matching or surpassing it at 22 $\times$ lower training cost.

Problem

• sparse novel view synthesis
• 3D geometry recovery
• view ambiguity
• geometric inconsistency
• floating artifacts

Innovation

• Visible Domain Constraint
• Sparse Novel View Synthesis
• Geometric Prior
• Silhouette Consistency
• Multi-view Visibility