This repo contains the training code and demo for NDE.

• python 3.8
• CUDA 11.7
• pytorch 2.0.1
• pytorch-lightning 2.0.8
• nerfacc
• tinycudann (fp32)

We compile tinycudann with fp32 precision for stable optimization. This is done by set TCNN_HALF_PRECISION=0 in this line.

• NeRF-Synthetic (Materials)
• RefNeRF-Synthetic
• NeRO-GlossyReal
• Teaser

The pre-trained weights for both synthetic and real scenes can be found in here

1. Edit configs/synthetic.yaml or configs/real.yaml to set up dataset path and configure a training.
2. To train a model, run:

python train.py --experiment_name=EXPERIMENT_NAME --device=GPU_DEVICE\
                --config CONFIG_FILE --max_epochs=NUM_OF_EPOCHS # 4000 by default

3. For view synthesis results, see demo/demo.ipynb

@inproceedings{wu2024neural,
  author = {Liwen Wu and Sai Bi and Zexiang Xu and Fujun Luan and Kai Zhang and Iliyan Georgiev and Kalyan Sunkavalli and Ravi Ramamoorthi},
  title = {Neural Directional Encoding for Efficient and Accurate View-Dependent Appearance Modeling},
  booktitle = {CVPR},
  year = {2024}
}