Learning for 3D Vision: HW4

• Author: Santiago Arambulo
• AndrewID: sarambul

1. 3D Gaussian Splatting

1.1 3D Gaussian Rasterization

1.2 Training 3D Gaussian Representations

1.3 Extensions: Rendering Using Spherical Harmonics

Without View Dependent Colours (same as before)

With View Dependent Colours (using Spherical Harmonics)

Frame number	Without View Dependent Colors	With View Dependent Colors	Differences
5			There are new shadows in the left side of the chair, both on the back and on the seat
13			The chair looks a lot less brighter from this angle with the view dependent colors
18			The details at the front edge of the seat are less distinguisable from this angle under the view dependent rendering. Also, there is more contrast between the shadowed regions and the rest of the chair.

2. Diffusion-guided Optimization

2.1 SDS Loss + Image Optimization

Prompt	Without Guidance	With Guidance
a hamburger
a standing corgi dog
a surfer
Eren Yeager in titan form

2.2 Texture Map Optimization for Mesh

Prompt	a light blue cow	a cow with a shinning red head and a matte white body
Rendering

2.3 NeRF Optimization

2.4 Extensions: View-dependent text embedding