16-889: Learning for 3D Vision

Late Days

1.5 Volume Rendering

Volume Rendering Cube

Depth Map

2.2 Loss and training

Box center: (0.25, 0.25, 0.0)

Box side lengths: (2.00, 1.50, 1.50)

2.3 Visualization

3 Optimizing a Neural Radiance Field

Model Complexity - Hidden Units (128)

Model Complexity - Hidden Units (256)

4.1 View Dependence

With increased view dependence, we will be able to capture the variations/effects such as reflections, shadows better but the tradeoff is that it can start to overfit on the images and reduce the generalizabily of the model when generating novel views. It is because the model will try to learn most of the things from the given images and fail to generalize on concepts common to different objects.

4.3 High Resolution

Default Config

Hidden Units (256)

Num Points Per Ray (256)