CMU-16889 Learning for 3D Vision - HW3

1. Differentiable Volume Rendering

1.3. Ray sampling

1.4. Point sampling

1.5. Volume rendering

2. Optimizing a basic implicit volume

2.2. Loss and training

Box center: (0.25, 0.25, 0.00)
Box side lengths: (2.00, 1.50, 1.50)

2.3. Visualization

The left image is my result. The right image is TA's result.

3. Optimizing a Neural Radiance Field (NeRF)

Visualization

4. NeRF Extras

4.1 View Dependence

By adding view dependence, the model may learn better reconstruction in the presence of specific views. The left image below shows the reconstruction of the model without view dependence. The right image shows the reconstruction with view dependence. We can see shadow inside the dipper of right image, while the color is relatively uniform in the left image. However, increasing view dependence might decrease generalization quality. Since the model is not a generative model, and thus does not have interpolation ability, it can be hard to generate good reconstruction from novel views.

4.3 High Resolution Imagery