Advancements in Gaussian Splatting for 3D Reconstruction

The field of 3D reconstruction is witnessing significant advancements with the development of Gaussian Splatting techniques. Researchers are exploring innovative methods to improve the accuracy and efficiency of 3D reconstruction in various environments, including underwater and aerial settings. The integration of semantic guidance, geometric constraints, and language models is leading to more robust and high-fidelity reconstructions. Notably, the use of multimodal cross-knowledge and visibility-aware language aggregation is enhancing the performance of Gaussian Splatting. Furthermore, the development of novel training strategies and object-level guidance is enabling the application of Gaussian Splatting in large scenes and complex environments. Overall, these advancements are pushing the boundaries of 3D reconstruction and opening up new possibilities for applications in fields such as underwater exploration, aerial robotics, and computer vision. Noteworthy papers include: SWAGSplatting, which proposes a novel framework for semantic-guided 3D Gaussian Splatting, and GeoSplat, which introduces a geometry-constrained optimization framework for Gaussian Splatting. Additionally, VIM-GS presents a monocular Gaussian Splatting framework using object-level guidance, and Visibility-Aware Language Aggregation improves open-vocabulary localization and segmentation in 3D Gaussian Splatting.

Sources

SWAGSplatting: Semantic-guided Water-scene Augmented Gaussian Splatting

CoRe-GS: Coarse-to-Refined Gaussian Splatting with Semantic Object Focus

GeoSplat: A Deep Dive into Geometry-Constrained Gaussian Splatting

Visibility-Aware Language Aggregation for Open-Vocabulary Segmentation in 3D Gaussian Splatting

VIM-GS: Visual-Inertial Monocular Gaussian Splatting via Object-level Guidance in Large Scenes

Built with on top of