MPGS: Multi-Plane Gaussian Splatting for Compact Scenes Rendering.

Area of Science:

• Computer Vision
• Computer Graphics
• Virtual and Augmented Reality

Background:

• High-fidelity rendering of heterogeneous scenes is challenging for Virtual Reality (VR) and Augmented Reality (AR).
• Existing 3D Gaussian Splatting (3DGS) methods struggle with weak-textured regions, producing artifacts and redundant data.

Purpose of the Study:

• To propose a novel Multi-Plane Gaussian Splatting (MPGS) method for high-fidelity rendering and compact reconstruction of heterogeneous scenes.
• To address limitations of current 3DGS in handling weak-textured areas and redundant Gaussian representations.

Main Methods:

• Introduced a multi-plane Gaussian optimization strategy to adapt Gaussian distributions for varying texture densities.
• Implemented a multi-scale geometric correction mechanism to prevent degradation of Gaussian distributions.
• Utilized normal information from compact scene learning for regularization of Gaussian distributions.

Main Results:

• MPGS demonstrated superior rendering quality compared to existing methods on public datasets.
• Achieved more compact scene reconstruction with reduced storage requirements.
• Offered more efficient rendering performance.

Conclusions:

• MPGS represents a state-of-the-art approach for compact reconstruction of heterogeneous scenes.
• The method significantly enhances rendering quality in novel view synthesis, particularly for weak-textured regions.
• MPGS enables more efficient and high-fidelity scene rendering for VR/AR applications.