https://github.com/hayden-donnelly/no-cuda-ngp-nerf

An educational implementation of Instant NGP NeRF using JAX and Numba instead of custom CUDA kernels.
https://github.com/hayden-donnelly/no-cuda-ngp-nerf

instant-ngp jax nerf neural-radiance-field numba

Last synced: about 10 hours ago
JSON representation

An educational implementation of Instant NGP NeRF using JAX and Numba instead of custom CUDA kernels.

• Host: GitHub
• URL: https://github.com/hayden-donnelly/no-cuda-ngp-nerf
• Owner: hayden-donnelly
• License: apache-2.0
• Created: 2023-11-04T00:00:43.000Z (about 1 year ago)
• Default Branch: main
• Last Pushed: 2024-06-07T03:36:11.000Z (6 months ago)
• Last Synced: 2024-06-07T04:32:03.328Z (6 months ago)
• Topics: instant-ngp, jax, nerf, neural-radiance-field, numba
• Language: Python
• Homepage:
• Size: 53.7 KB
• Stars: 2
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE.txt

Awesome Lists containing this project

README

        
# no-cuda-ngp-nerf

An educational implementation of Instant NGP NeRF using JAX and Numba instead of custom CUDA kernels.

Currently a work-in-progress.

## Notes 

- Checkout [derivation_of_derivatives.md](./derivation_of_derivatives.md) to see how the volume rendering derivatives are derived.

## Installation

First clone the repository.

```

https://github.com/hayden-donnelly/no-cuda-ngp-nerf.git

```

Then cd into it and install with pip.

```

cd no-cuda-ngp-nerf

```

```

python3 -m pip install -e .

```

The ``-e`` flag will let you edit the project without having to reinstall.

## Development Environment 

### Docker

- To build the docker image, use `docker-compose build`.

- To start the docker container, use `docker-compose up -d`.

- To open a shell inside the container, use `docker-compose exec ngp bash`.

- To open Jupyter Lab inside the container instead of a shell, go to [http://localhost:7070/](http://localhost:7070/).

- To stop the container, use `docker-compose down`.

### Nix

TODO

## TODO

- [ ] Switch to a Nix development environment.

- [ ] Add occupancy grid bitfield.

- [ ] Add screenshot/video of a trained NeRF render.

## Acknowledgement

Kwea123's [video lecture](https://www.youtube.com/live/c2t_C4-Ovss?si=PYRWj1IZP5y0nJms) and PyTorch implementation of [Instant NGP](https://github.com/kwea123/ngp_pl) were both very helpful during the development of this project.

## Citation

```bibtex

@article{mueller2022instant,

    author={Thomas M\"uller and Alex Evans and Christoph Schied and Alexander Keller},

    title={Instant Neural Graphics Primitives with a Multiresolution Hash Encoding},

    journal={ACM Trans. Graph.},

    issue_date={July 2022},

    volume={41},

    number={4},

    month=jul,

    year={2022},

    pages={102:1--102:15},

    articleno={102},

    numpages={15},

    url={https://doi.org/10.1145/3528223.3530127},

    doi={10.1145/3528223.3530127},

    publisher={ACM},

    address={New York, NY, USA},

}

```