A Survey of 3D Gaussian Splatting: Optimization Techniques, Applications, and AI-Driven Advancements

Abstract

3D Gaussian Splatting (3DGS) has emerged as a transformative technique for real-time, high-quality 3D rendering. This paper surveys the advancements in 3DGS, focusing on optimization strategies, applications across diverse domains, and AI-driven innovations that enhance its utility. Optimization techniques, such as Mini-Splatting and sparsification frameworks, address challenges like computational overhead and memory inefficiency while maintaining visual fidelity. AI has further expanded the capabilities of 3DGS through frameworks like SceneTeller, ART3D and GaussianEditor, enabling text-to-3D generation, object-level manipulation, and artistic scene creation with multi-view consistency. Applications of 3DGS range from autonomous driving and robotics to virtual and augmented reality, highlighting its adaptability and efficiency. This survey underscores the importance of continued research to refine 3DGS’s scalability and applicability, solidifying its role at the intersection of computer graphics and artificial intelligence.