Reconstruction and generation of 3D scenes
-
Neural fields for 3D scene reconstruction
In this lecture, we will discuss the emerging area of neural fields and their application to 3D scene reconstruction. We will discuss the underlying volume rendering algorithm, a variety of architectures for efficient rendering, and other solutions necessary for faithful reconstruction.
-
Neural fields for text-to-3D scene generation
Going beyond scene reconstruction, we will show how a pre-trained text-to-2d diffusion model can apply to text-to-3d scene generation. We will define the score distillation sampling loss and will give an overview of scene generation solution based on this technique.