Advances in 3D Reconstruction and Novel View Synthesis

The field of 3D reconstruction and novel view synthesis is rapidly advancing with the development of new methods and techniques. A key direction in this field is the use of Gaussian-based scene representations, which have been shown to be effective in achieving robust tracking and high-fidelity mapping. Another area of research is the development of methods for underwater scene reconstruction, which is a challenging task due to the limited quality of underwater images. Researchers are also exploring the use of Neural Radiance Fields (NeRFs) for large-scale scene modeling, which has shown promising results in terms of scalability and accuracy. Furthermore, there is a growing interest in developing methods for sparse view reconstruction, which can be used in applications such as robotic platforms. Noteworthy papers in this area include AquaGS, which presents an SfM-free underwater scene reconstruction model, and Switch-NeRF++, which introduces a Heterogeneous Mixture of Hash Experts network for scalable NeRFs. Additionally, SparSplat presents a fast multi-view reconstruction method with generalizable 2D Gaussian Splatting, and GSsplat proposes a generalizable semantic Gaussian Splatting method for efficient novel-view synthesis.

Sources

Unconstrained Large-scale 3D Reconstruction and Rendering across Altitudes

High Dynamic Range Novel View Synthesis with Single Exposure

GauS-SLAM: Dense RGB-D SLAM with Gaussian Surfels

AquaGS: Fast Underwater Scene Reconstruction with SfM-Free Gaussian Splatting

Visual enhancement and 3D representation for underwater scenes: a review

Learning Heterogeneous Mixture of Scene Experts for Large-scale Neural Radiance Fields

SparSplat: Fast Multi-View Reconstruction with Generalizable 2D Gaussian Splatting

Holographic Radiance Cascades for 2D Global Illumination

Sparfels: Fast Reconstruction from Sparse Unposed Imagery

Sparse Ellipsoidal Radial Basis Function Network for Point Cloud Surface Representation

3D Surface Reconstruction with Enhanced High-Frequency Details

AquaticVision: Benchmarking Visual SLAM in Underwater Environment with Events and Frames

Scalable Aerial GNSS Localization for Marine Robots

GSsplat: Generalizable Semantic Gaussian Splatting for Novel-view Synthesis in 3D Scenes

SGCR: Spherical Gaussians for Efficient 3D Curve Reconstruction

CPP-DIP: Multi-objective Coverage Path Planning for MAVs in Dispersed and Irregular Plantations

Time of the Flight of the Gaussians: Optimizing Depth Indirectly in Dynamic Radiance Fields

Built with on top of