16-889: Learning for 3D Vision (Spring 2022)

Assignmenet 3

Qichen Fu

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

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)

Explainations:
The increased view dependence allows the model to give different feature from different view directions. By doing so, the model can create more realistic effects such like the specular reflection.
However, increased the view dependence increases the model complexity since the input channel of the MLP is increased. Meanwhile, it needs more images from various view points convering each single 3D point to give the view dependence supervision signal. Otherwise, the model may overfit to a specific view. So it limited the model's generalization quality under the condition of limited data (images from different views).

4.3 High Resolution Imagery (10 pts)

The hyper parameters are shown in the config and code files submitted to canvas.