16825 - HW3

This is a sample paragraph where you can describe your project, experiment, or any text content you want. You can add more paragraphs by copying this block.

0. Transmittance Calculation

1.1 - 1.4 Grid, Ray, Point Sampling

1.5 Volume Rendering

2. Optimizing implicit volume

3. NeRF

4. View dependence

While view dependence enables complex optical effects like reflections, I observed aliasing in the rendered images. This might be caused by the neural network is bigger, thus becoming harder to train.

5. Sphere Tracing

6. Optimizing Neural SDF

7. Vol SDF

  1. Beta controls the Lipschitz continuity of the density function. A high beta makes density change slower, making geometry harder to fit to thin structures. A low beta allows sharp changes in density, thus enabling better fitting to thin structures, but may introduce random noise.
  2. It would be easier to train with low beta as it relaxes the density function (allowing it to change faster).
  3. Better learn accurate surfaces with low beta, but may have noise in the volume, given sufficient data.

8. Complex Scene