Advances in Gaussian Splatting for 3D Reconstruction and Scene Understanding

The field of computer vision is witnessing significant advancements in 3D reconstruction and scene understanding, driven by innovations in Gaussian Splatting. Recent developments have focused on addressing challenges such as temporal misalignment, sparse-view reconstruction, and robustness to occlusion. Researchers are exploring novel approaches to integrate Gaussian Splatting with other techniques, such as diffusion models and neural rendering, to enhance reconstruction quality and efficiency. Notable papers in this area include Dynamic Gaussian Scene Reconstruction from Unsynchronized Videos, which proposes a temporal alignment strategy for high-quality 4DGS reconstruction from unsynchronized multi-view videos, and SRSplat, which introduces a feed-forward framework for reconstructing high-resolution 3D scenes from sparse, low-resolution images. Other noteworthy papers include LiDAR-GS++, which enhances LiDAR Gaussian Splatting reconstruction using diffusion priors, and iGaussian, which achieves real-time camera pose estimation via feed-forward 3D Gaussian Splatting inversion.

Sources

Dynamic Gaussian Scene Reconstruction from Unsynchronized Videos

SRSplat: Feed-Forward Super-Resolution Gaussian Splatting from Sparse Multi-View Images

LiDAR-GS++:Improving LiDAR Gaussian Reconstruction via Diffusion Priors

SplatSearch: Instance Image Goal Navigation for Mobile Robots using 3D Gaussian Splatting and Diffusion Models

iGaussian: Real-Time Camera Pose Estimation via Feed-Forward 3D Gaussian Splatting Inversion

Gaussian Splatting-based Low-Rank Tensor Representation for Multi-Dimensional Image Recovery

Iterative Diffusion-Refined Neural Attenuation Fields for Multi-Source Stationary CT Reconstruction: NAF Meets Diffusion Model

Rad-GS: Radar-Vision Integration for 3D Gaussian Splatting SLAM in Outdoor Environments

Clustered Error Correction with Grouped 4D Gaussian Splatting

How Robot Dogs See the Unseeable

CRISTAL: Real-time Camera Registration in Static LiDAR Scans using Neural Rendering

EOGS++: Earth Observation Gaussian Splatting with Internal Camera Refinement and Direct Panchromatic Rendering

Built with on top of