Advances in 3D Gaussian Splatting for Novel View Synthesis and Scene Reconstruction

The field of 3D Gaussian Splatting is rapidly advancing, with a focus on improving novel view synthesis and scene reconstruction. Recent developments have led to the creation of more efficient and effective methods for generating high-quality 3D scenes from incomplete observations. Notably, researchers are exploring the use of geometric priors, continuous level-of-detail techniques, and robust pose-free reconstruction methods to enhance the accuracy and realism of 3D scenes. Additionally, there is a growing interest in integrating 3D Gaussian Splatting with other techniques, such as signed distance functions and learned image priors, to further improve rendering quality and reduce storage overhead. Overall, these advancements have significant implications for applications in computer vision, robotics, and augmented reality. Noteworthy papers include Geometry-Aware Scene Configurations for Novel View Synthesis, which proposes scene-adaptive strategies for efficient representation capacity allocation, and CLoD-GS, which introduces a continuous level-of-detail mechanism for smooth quality scaling. VA-GS is also notable for its view alignment method, which enhances the geometric representation of 3D Gaussians. UniGS and PAGS demonstrate the potential of unified geometry-aware Gaussian Splatting for multimodal rendering and priority-adaptive Gaussian Splatting for dynamic driving scenes, respectively.

Sources

Geometry-Aware Scene Configurations for Novel View Synthesis

CLoD-GS: Continuous Level-of-Detail via 3D Gaussian Splatting

Gesplat: Robust Pose-Free 3D Reconstruction via Geometry-Guided Gaussian Splatting

Dynamic Gaussian Splatting from Defocused and Motion-blurred Monocular Videos

MaterialRefGS: Reflective Gaussian Splatting with Multi-view Consistent Material Inference

VA-GS: Enhancing the Geometric Representation of Gaussian Splatting via View Alignment

UniGS: Unified Geometry-Aware Gaussian Splatting for Multimodal Rendering

PAGS: Priority-Adaptive Gaussian Splatting for Dynamic Driving Scenes

BSGS: Bi-stage 3D Gaussian Splatting for Camera Motion Deblurring

SimULi: Real-Time LiDAR and Camera Simulation with Unscented Transforms

Leveraging 2D Priors and SDF Guidance for Dynamic Urban Scene Rendering

Beyond Pixels: A Differentiable Pipeline for Probing Neuronal Selectivity in 3D

BalanceGS: Algorithm-System Co-design for Efficient 3D Gaussian Splatting Training on GPU

Leveraging Learned Image Prior for 3D Gaussian Compression

Built with on top of