16-889 Assignment 3: Volume Rendering and Neural Radiance Fields

1. Differentiable Volume Rendering

1.3. Ray sampling (10 points)

1.4. Point sampling (10 points)

1.5. Volume rendering (30 points)

2. Optimizing a basic implicit volume

2.2. Loss and training (5 points)

Box center: (0.24981717765331268, 0.2501944899559021, -0.000944553641602397)

Box side lengths: (2.003711223602295, 1.5006524324417114, 1.5030611753463745)

2.3. Visualization

3. Optimizing a Neural Radiance Field (NeRF) (30 points)

4. NeRF Extras (Choose at least one! More than one is extra credit)

4.1 View Dependence (10 pts)

If we increase view dependence, NeRF may be able to learn photorealistic view-dependent reflection. However, if the frequency bands of the positional encoding are large, NeRF may lose the ability of view-dependent interpolation since the slight view difference changes the encoded feature significantly. Therefore, we need to be careful about the frequency bands when increasing view dependence.

4.3 High Resolution Imagery (10 pts)

Left: Original, Right: Original with view dependence