A Survey on 3D Gaussian Splatting Applications: Segmentation, Editing, and Generation

Area of Science:

• Computer Vision
• Computer Graphics
• Artificial Intelligence

Background:

• 3D Gaussian Splatting (3DGS) is a novel technique for real-time, high-fidelity novel view synthesis.
• It offers an explicit and compact representation, distinguishing it from Neural Radiance Fields (NeRF).
• Its capabilities extend beyond rendering to tasks requiring geometric and semantic understanding.

Purpose of the Study:

• To provide a comprehensive survey of recent advancements in 3D Gaussian Splatting (3DGS) applications.
• To categorize and analyze downstream tasks enabled by 3DGS.
• To highlight emerging trends and provide resources for further research.

Main Methods:

• Review of 3DGS reconstruction preliminaries and problem formulation.
• Exploration of 2D foundation models and NeRF-based research relevant to 3DGS.
• Categorization of 3DGS applications into foundational tasks (segmentation, editing, generation) and functional applications.

Main Results:

• Summaries of representative methods, supervision strategies, and learning paradigms for 3DGS applications.
• Identification of shared design principles and emerging trends across different application areas.
• Comparative analysis of recent methods using common datasets and evaluation protocols.

Conclusions:

• 3DGS is a versatile technique with significant potential in various computer vision and graphics applications.
• The survey categorizes applications and identifies key trends, facilitating future research.
• A continuously updated repository of resources is provided to support the research community.