16-825 Assignment 3

A. Neural Volume Rendering

1.3. Ray sampling

1.4. Point sampling

1.5. Volume rendering

2.2. Loss and training

Box center: (0.25, 0.25, 0.00)

Box side lengths: (2.01, 1.50, 1.50)

2.3. Visualization

3. Optimizing a Neural Radiance Field (NeRF)

4.1. View Dependence

The biggest advantage is that the objects look more realistic, and light bounces off them more realistically. The risk is that the network can cheat and learn each training view's colours through the direction input instead of understanding the geometry.

B. 3D Gaussian Splatting

5. 3D Gaussian Rasterization

6. Training 3D Gaussian Representations

Learning rates: opacities 0.05, scales 0.005, colours 0.02, means 0.0005

Iterations: 1000

PSNR: 29.83

SSIM: 0.940

7.1. Rendering Using Spherical Harmonics

Without spherical harmonics (5.5):

With spherical harmonics (7.1):

Side-by-side comparisons (left: without spherical harmonics, right: with spherical harmonics):

With spherical harmonics, the green fabric looks darker with more contrast, while without them it looks flat and evenly lit.

The seat and back show richer, more realistic shading with spherical harmonics, since colour now changes with viewing direction.

The underside looks almost the same in both, since it has little view-dependent lighting and the base colour is already close.