3D Gaussian Splatting | Guest Lecture for "Computer Graphics in AI Era"

This is a recording of my guest lecture for CS8803/4803 CGA -- "Computer Graphics in AI Era", a Georgia Tech course taught by Prof. Bo Zhu (https://faculty.cc.gatech.edu/~bozhu/ ).

With students ranging from undergrads with minimal background to master's students experienced in computer vision, machine learning, and graphics, my goal was to make the lecture engaging and useful for everyone.

🌐 Course website: https://cgai-gatech.vercel.app
🖼️ "Tiny gaussian splatting in 2D" notebook: https://github.com/bobarna/tiny-2d-ga...

📅 Recorded on: Feb. 27, 2025
---
⏳ Timestamps:
00:00 Overview
01:30 Differentiable Models in CG / The Big Picture
03:47 Novel View Synthesis
04:45 Recap: Neural Radiance Fields (NeRF)
07:40 Two examples: Differentiable Rendering in 2D
09:10 3D Gaussian Splatting Demo [1]
11:20 3D Gaussian Splatting vs NeRF
12:49 3DGS Pipeline Overview
14:52 Parametrization: 1D/2D/3D Gaussians
16:17 Parametrization: 2D Gaussian, Building the Sigma Matrix
24:33 Initialization: Structure from Motion (SfM)
31:23 Spherical Harmonics (SH) quick overview
35:29 3D-to-2D Projection
38:50 2D Rasterization (alpha blending)
49:39 Optimization
56:50 Anisotropic Gaussians
57:35 "4D Gaussian Splatting" [2]
58:55 Next Steps [3]
1:01:19 Fun fact: Fourier features

📺 My second guest lecture on Differentiable Physics:    • Differentiable Physics | Guest Lecture for...  


[1]: https://vincentwoo.com/3d/sutro_tower
[2]: For details, see the "Dynamic 3D Gaussians: Tracking by Persistent Dynamic View Synthesis" paper: https://dynamic3dgaussians.github.io/

[3]
Also see:
3DGS paper: https://repo-sam.inria.fr/fungraph/3d...
2D Gaussian Splatting: https://surfsplatting.github.io/
TUM AI Lecture Series - The 3D Gaussian Splatting Adventure: Past, Present, Future (George Drettakis): https://www.youtube.com/live/DjOqkVIlEGY

Trying 3DGS at home:
gsplat: https://docs.gsplat.studio/main
NeRF studio: https://docs.nerf.studio/