CylinderSplat: 3D Gaussian Splatting with Cylindrical Triplanes for Panoramic Novel View Synthesis

2026-03-06

AI Summary

Existing 3D Gaussian splatting methods struggle with occlusion handling in novel view synthesis under sparse viewpoints, and conventional Cartesian triplane representations fail to effectively model the geometry of 360° scenes, often leading to distortion and aliasing. This work proposes a feedforward panoramic 3D Gaussian splatting framework that introduces, for the first time, a cylindrical triplane representation aligned with the Manhattan world assumption. The architecture features a dual-branch design: a pixel branch reconstructs visible regions, while a voxel branch leverages the cylindrical triplane to complete occluded or sparsely observed areas. By seamlessly integrating local detail with global structure, the method achieves state-of-the-art performance in both single-view and multi-view panoramic novel view synthesis, significantly improving reconstruction quality and geometric accuracy.

Technology Category

• Computer Vision: 3D Computer Vision
• Intelligent Robots: Multimodal Perception & Sensor Fusion
• Machine Learning: Learning with Manifolds

Application Category

• Economics, Online Markets and Human Computation: Architectures and workflows that use LLMs for crowd work
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Feed-forward 3D Gaussian Splatting (3DGS) has shown great promise for real-time novel view synthesis, but its application to panoramic imagery remains challenging. Existing methods often rely on multi-view cost volumes for geometric refinement, which struggle to resolve occlusions in sparse-view scenarios. Furthermore, standard volumetric representations like Cartesian Triplanes are poor in capturing the inherent geometry of $360^\circ$ scenes, leading to distortion and aliasing. In this work, we introduce CylinderSplat, a feed-forward framework for panoramic 3DGS that addresses these limitations. The core of our method is a new {cylindrical Triplane} representation, which is better aligned with panoramic data and real-world structures adhering to the Manhattan-world assumption. We use a dual-branch architecture: a pixel-based branch reconstructs well-observed regions, while a volume-based branch leverages the cylindrical Triplane to complete occluded or sparsely-viewed areas. Our framework is designed to flexibly handle a variable number of input views, from single to multiple panoramas. Extensive experiments demonstrate that CylinderSplat achieves state-of-the-art results in both single-view and multi-view panoramic novel view synthesis, outperforming previous methods in both reconstruction quality and geometric accuracy.

Problem

• panoramic novel view synthesis
• 3D Gaussian Splatting
• occlusion handling
• cylindrical representation
• sparse-view reconstruction

Innovation

• Cylindrical Triplane
• 3D Gaussian Splatting
• Panoramic Novel View Synthesis
• Manhattan-world assumption
• Feed-forward Framework