This repository contains Yumao Liu's personal learning notes and practical explorations of two state-of-the-art scene representation learning techniques:
- Neural Radiance Fields (NeRF)
- Gaussian Splatting (GS)
The goal is to develop a deep, hands-on understanding of both the mathematics and implementation details behind modern neural scene representation and rendering.
NeRF vs 3D Gaussian Splatting
-
nerf_from_nothing/:
My self-written notebook and explanatory materials on NeRF, including core theory, code, visualizations, and step-by-step explanations of key mechanisms such as ray sampling, volume rendering, positional encoding, and hierarchical sampling. -
gaussian_splatting/:
My structured English notes summarizing the core ideas and algorithmic steps behind Gaussian Splatting, including initialization from SfM points, 3D Gaussian formulation, image-space projection, differentiable rasterization, loss computation, and adaptive density control.
- Demonstrates hands-on understanding of leading neural representation learning methods, not just from papers but also through working code and visualization.
- Serves as a reference for others interested in mastering modern 3D scene modeling pipelines.
- Complements my research experience in SLAM, 3D vision, and perception for robotics.
All original code and notes released under the MIT License. For any third-party materials, see individual folders for details.