PanoPlane: Plane-Aware Panoramic Completion for Sparse-View Indoor 3D Gaussian Splatting

2026-05-15 · Source: cs.CV updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Computer Vision, Robotics & Autonomous Systems · Depth: Expert, extended

Summary

PanoPlane is a novel approach for high-fidelity sparse-view indoor novel view synthesis that reconstructs closed room geometry using panoramic scene completion. Unlike methods that rely on limited perspective views, PanoPlane leverages $360^{\circ}$ panoramic completion to condition its generative process on the full spatial layout. It introduces Layout Anchored Attention Steering, a training-free mechanism that guides a diffusion model's internal attention towards detected planar surfaces (walls, floors, ceilings) during inference. This mechanism replaces unconstrained hallucination with grounded surface extrapolation, providing globally consistent supervision for 3D Gaussian Splatting. The method achieves state-of-the-art novel view synthesis quality on Replica, ScanNet++, and Matterport3D datasets, demonstrating up to a +17.8% improvement in PSNR over current baselines with as few as three input views, without any training or fine-tuning of the diffusion model.

Key takeaway

For research scientists working on sparse-view 3D reconstruction, PanoPlane demonstrates that integrating panoramic scene completion with 3D planar structure guidance significantly enhances geometric consistency and novel view synthesis quality. You should consider adopting panoramic representations and attention steering techniques to improve reconstruction accuracy, especially in indoor environments with dominant planar surfaces, as this approach yields superior results without requiring extensive model retraining.

Key insights

Panoramic scene completion with layout-anchored attention steering significantly improves sparse-view 3D reconstruction.

Principles

• Panoramic views offer richer global context than perspective images.
• Anchoring generative models to 3D planar structure improves geometric consistency.
• Attention steering can enhance diffusion model performance without retraining.

Method

PanoPlane initializes a coarse 2DGS, detects and classifies planar surfaces, renders partial panoramas, assigns unobserved regions to layout planes via geometry and boundary-based methods, and then uses layout-anchored attention steering during panoramic flow-matching inpainting.

In practice

• Use panoramic representations for global scene context.
• Classify planes semantically to guide geometric steering.
• Combine geometric and boundary assignments for robust hole filling.

Topics

• PanoPlane
• 3D Gaussian Splatting (3DGS)
• Sparse-View Novel View Synthesis
• Panoramic Scene Completion
• Layout Anchored Attention Steering

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

Related on AIssential

• PanoImager: Geometry-Guided Novel View Synthesis and Reconstruction from Sparse Panoramic Views — PanoImager enables robust 3D reconstruction from sparse panoramas by integrating pose/depth priors, diffusion view completion, and 3DGS optimization.
• Rein3D: Reinforced 3D Indoor Scene Generation with Panoramic Video Diffusion Models — Rein3D integrates 3D Gaussian Splatting with panoramic video diffusion to restore and refine 3D indoor scenes from sparse inputs.
• GlobalSplat: Efficient Feed-Forward 3D Gaussian Splatting via Global Scene Tokens — GlobalSplat uses a global latent scene representation for efficient, compact, and consistent 3D Gaussian Splatting.
• Deep Learning Approaches for 3D Medical Scene Completion: From Geometric Modeling to Generative Paradigms — The field of 3D medical scene completion has rapidly evolved from voxel-based methods to generative diffusion and Gaussian splatting.
• FLUX3D: High-Fidelity 3D Gaussian Generation with Diffusion-Aligned Sparse Representation — FLUX3D enhances 3D Gaussian Splatting generation fidelity by resolving representation and cross-modal alignment bottlenecks.
• Feed-Forward 3D Scene Modeling: A Problem-Driven Perspective — Feed-forward 3D reconstruction offers efficient, generalizable scene modeling by directly mapping 2D inputs to 3D representations.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.CV updates on arXiv.org.