Normal-NeRF is built on nerfstudio. Follow their installation guide to create the environment and install dependencies.
Once you have finished installing nerfstudio, you can install Normal-NeRF using the following commands:
git clone https://github.com/sjj118/Normal-NeRF.git
cd Normal-NeRF
pip install ./griddecay
pip install -e .
Download datasets from following links: NeRF Synthetic, Shiny Blender and Glossy Synthetic.
To preprocess the Shiny Blender dataset, run:
python preprocess_shiny.py data/shiny_blender
For Glossy Synthetic dataset, download both GlossySynthetic.tar.gz and glossy-synthetic-nvs.zip, then extract their contents into the same directory. For example:
GlossySynthetic
├── angel
├── angel_nvs
├── bell
├── bell_nvs
├── ...
Next, convert the dataset format by running the following command:
python preprocess_glossy.py data/GlossySynthetic data/glossy_synthetic
# Training
ns-train normalnerf --data data/nerf_synthetic/ship
# Evaluation
ns-eval --load-config <path to experiment's config.yml>
# Rendering
ns-render dataset --image-format png --load-config <path to experiment's config.yml>