3D Gaussian Splatting, while being primarily a novel view synthesis method, is adapted for 3D surface reconstruction by methods such as SuGaR. Some methods try to reconstruct the surface as an ad-hoc method while some train a traditional neural SDF in conjunction with the 3D Gaussians.
In this project, the goal is to try to construct primitives such that surface can obtained from 3D Gaussian scene with minimal extra computational cost.
Objective
- Literature Review of relevant methods
- Experimentation with different SDF formulations
Prerequisites
- Proficient with PyTorch and Python.
- Working knowledge of computer vision or computer graphics is recommended.
- Knowledge of Gaussian splatting method is a plus.
Contact
References
- 3D Gaussian Splatting
for Real-Time Radiance Field Rendering, SIGGRAPH 2023 - SuGaR: Surface-Aligned Gaussian Splatting for Efficient 3D Mesh Reconstruction and High-Quality Mesh Rendering, CVPR 2024
- Neuralangelo: High-Fidelity Neural Surface Reconstruction, CVPR 2023
- 3DGSR: Implicit Surface Reconstruction with 3D Gaussian Splatting