Bonus

This unit took a closer look at Gaussian Splatting, a differentiable rasterization technique that also serves as an ML-friendly 3D representation. While this is a special case due to its real-time rendering capabilities, it’s one of many representations used in generative 3D research.

Most recently, triplanes have emerged as the latest state-of-the-art in final mesh quality, used in pipelines like InstantMesh. You can learn more about them in this Community Notebook provided by Damian.

If you’re interested in learning more about Gaussian Splatting, here are some pointers:

• Nerfstudio gsplat: An open-source implementation of Gaussian Splatting
• gsplat.js: An open-source JavaScript library for splat rendering
• UnityGaussianSplatting: A Unity implementation of Gaussian Splatting (don’t be fooled by it being labeled a “toy” - it’s a powerful tool!)
• In the capstone of this course, you’ll have the option of targeting meshes or Gaussian Splatting as the final output of your generative model

In the next unit, we’ll be diving into meshes, the representation used in 3D applications everywhere.

< > Update on GitHub