ARM-HashNeRF-pytorch

This project implements Accelerated Ray Marching (ARM) in HashNerf-pytorch, a pure PyTorch implementation of Instant-NGP. Instant-NGP drastically reduces (up to two orders of magnitude) the cost of training and evaluation of Neural Graphics Primitives that are parametrized by fully connected neural networks.

ARM-HashNeRF vs Vanilla HashNeRF

Both the rendering time and quality of ARM-HashNeRF are compared against Vanilla HashNeRF. In all cases, rendering time is reduced while resulting in only a minimal decrease in redering quality (PSNR). For instance, in the 50K iterations comparison below, ARM-HashNeRF achieves 9.71% faster rendering compared to Vanilla HashNeRF, with only a 5.43% reduction in PSNR. Vanilla HashNeRF is on the left and ARM-HashNeRF on the right. All experiments were run using a single Tesla P100 GPU.

original-vs-arm_50K_default.mp4

Contents

• Instructions
• Citation
• Acknowledgments

Instructions

1. Download Dataset

The NeRF synthetic LEGO dataset is used in this project. Please download the preprocessed dataset from here and place it in the ARM-HashNeRF-pytorch/ directory.

2. Clone Repository

git clone git@github.com:jorgedanielrodrividal/ARM-HashNeRF-pytorch.git

3. Install custom vren library

pip install art/csrc/

4. Training

python run_arm_nerf.py --config configs/lego.txt --finest_res 512 --log2_hashmap_size 19 --lrate 0.01 --lrate_decay 10

---

Citation

This project is mostly based on the amazing work of:

@misc{bhalgat2022hashnerfpytorch,
  title={HashNeRF-pytorch},
  author={Yash Bhalgat},
  publisher = {GitHub},
  journal = {GitHub repository},
  howpublished={\url{https://github.com/yashbhalgat/HashNeRF-pytorch/}},
  year={2022}
}

@article{mueller2022instant,
    title = {Instant Neural Graphics Primitives with a Multiresolution Hash Encoding},
    author = {Thomas M\"uller and Alex Evans and Christoph Schied and Alexander Keller},
    journal = {arXiv:2201.05989},
    year = {2022},
    month = jan
}

If you find this work useful, feel free to cite:

@misc{jorgedaniel2025armhashnerfpytorch,
  title={ARM-HashNeRF-pytorch},
  author={Jorge Daniel},
  publisher = {GitHub},
  journal = {GitHub repository},
  howpublished={\url{https://github.com/jorgedanielrodrividal/ARM-HashNeRF-pytorch/}},
  year={2025}
}

---

Acknowledgments

Big thanks to Yash Bhalgat for his enlightening HashNeRF project. Also thanks to the author of ngp_pl, which served as a key inspiration for the ARM implementation.