Overview

Summary:

• Uses two NeRF models: one for background and one for foreground.
• The background NeRF model focuses on capturing the environment of the scene.
• The foreground NeRF model is responsible for capturing the main objects in the scene.
• Partitions the scene space into two volumes: an inner unit sphere (foreground) and an outer volume (background) represented by an inverted sphere covering the complement of the inner volume.

Poses and camera parameters: Obtained using Metashape.

Images: Used lf idol dataset.

Drawbacks: Requires large amount of data, slow to train.

Paper: NeRF++ paper

Results:: The below results are for a NeRF++ model trained on the lf idol dataset.