VisDom: Sparse Novel View Synthesis with Visible Domain Constraint

2026-06-18 · Source: Computer Vision and Pattern Recognition · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems · Depth: Expert, quick

Summary

VisDom is a novel, learning-free geometric constraint designed to enhance sparse novel view synthesis (NVS) by addressing overfitting issues common in NeRF- and Gaussian Splatting (GS)-based methods when few input views are available. This technique augments classical carving-based visual hull reconstruction by enforcing a minimum multi-view visibility requirement. Specifically, VisDom defines a "visible domain" as the subset of 3D space observed by at least K views, applying this as an additional filtering criterion alongside standard silhouette-based reconstruction. It integrates seamlessly into both implicit (NeRF) and explicit (GS) pipelines, restricting volumetric sampling and guiding Gaussian placement during optimization. Experimental results across three challenging datasets demonstrate consistent improvements in sparse-view NVS, facilitating high-quality object-centric reconstruction from as few as four input images. VisDom is domain-agnostic, requires only silhouettes, and introduces no learned parameters, making it a simple, effective complement to existing approaches, even improving GaussianObject performance on Omni3D and MipNeRF360 at 22x lower training cost.

Key takeaway

For computer vision engineers developing sparse novel view synthesis systems, consider integrating VisDom to significantly improve reconstruction quality from limited input. This learning-free geometric constraint effectively mitigates overfitting and floating artifacts, enabling high-fidelity object-centric models with as few as four images. You can seamlessly apply it to both NeRF and Gaussian Splatting pipelines, potentially reducing training costs by up to 22 times while matching or exceeding current performance benchmarks.

Key insights

VisDom uses a multi-view visibility constraint to improve sparse novel view synthesis by reducing overfitting and artifacts.

Principles

• Sparse NVS benefits from strong spatial priors.
• Silhouette consistency alone is often insufficient.
• Learning-free geometric constraints can enhance NVS.

Method

VisDom defines a visible domain as 3D space seen by ≥K views, applying it as a filter with silhouette-based reconstruction to restrict sampling and guide Gaussian placement.

In practice

• Reconstruct objects from as few as four images.
• Integrate into existing NeRF or GS pipelines.
• Reduce training cost for NVS models significantly.

Topics

• Sparse Novel View Synthesis
• NeRF
• Gaussian Splatting
• 3D Reconstruction
• Visible Domain Constraint
• Multi-view Geometry

Best for: Research Scientist, AI Scientist, Computer Vision Engineer

Related on AIssential

• Fast and Lightweight Novel View Synthesis with Differentiable Multiplane Image — MPI, enhanced with diffusion and point maps, offers a fast, lightweight alternative for novel view synthesis.
• UniSHARP: Universal Sharp Monocular View Synthesis — UniSHARP enables universal monocular novel view synthesis across diverse camera types using a unified ray-distance Gaussian representation.
• Large Depth Completion Model from Sparse Observations — LDCM achieves robust, metric-accurate dense depth completion from sparse observations via a transformer, novel initialization, and 3D coordinate regression.
• DF3DV-1K: A Large-Scale Dataset and Benchmark for Distractor-Free Novel View Synthesis — DF3DV-1K is a new dataset for benchmarking distractor-free novel view synthesis methods.
• Resolving Representation Ambiguity in Feedforward Novel View Synthesis Transformer via Semantic-Spatial Decoupling — Decoupling semantic and spatial representations in NVS transformers improves rendering fidelity by preventing spatial bias.
• PanoPlane: Plane-Aware Panoramic Completion for Sparse-View Indoor 3D Gaussian Splatting — Panoramic scene completion with layout-anchored attention steering significantly improves sparse-view 3D reconstruction.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Computer Vision and Pattern Recognition.