PanoImager: Geometry-Guided Novel View Synthesis and Reconstruction from Sparse Panoramic Views

2026-06-25

AI Summary

This work addresses the challenge of unreliable initialization in traditional Structure-from-Motion (SfM) and SLAM methods under rotation-dominated, sparsely sampled panoramic imagery with weak parallax, which often leads to failure in 3D reconstruction. To overcome this limitation, the authors propose an end-to-end framework that bypasses SfM entirely: it first estimates initial camera poses and depth maps using a feedforward network, then completes missing viewpoints via a geometry-conditioned diffusion model, and finally refines the reconstruction through depth-guided 3D Gaussian splatting. This approach represents the first integration of geometry-guided diffusion models with 3D Gaussian splatting, enabling highly consistent novel view synthesis and stable reconstruction even from extremely sparse inputs. Experiments demonstrate that the method significantly outperforms existing approaches across multiple benchmarks and can serve as a robust map refinement module when SfM or SLAM fails.

Abstract

Panoramic sensing offers wide field-of-view coverage, yet 3D reconstruction from sparse panoramas remains challenging under rotation-dominant, weak-parallax motion. In such regimes, SfM/SLAM initialization is often ill-conditioned and unreliable. We present PanoImager, an SfM-free framework that combines feed-forward pose/depth priors, geometry-conditioned diffusion view completion, and depth-guided 3DGS optimization. Given only a few panoramic images, PanoImager decomposes them into local perspective views, synthesizes auxiliary observations to enrich sparse evidence, and stabilizes Gaussian optimization for improved cross-view consistency. Experiments on multiple benchmarks show improved stability under extreme sparsity, suggesting PanoImager as an offline/background component for map refinement when SfM/SLAM fails to initialize.

Problem

• 3D reconstruction
• sparse panoramic views
• weak-parallax motion
• SfM/SLAM initialization
• novel view synthesis

Innovation

• geometry-guided diffusion
• sparse panoramic reconstruction
• SfM-free
• depth-guided 3DGS
• novel view synthesis