Neural Radiance Field Research Repository

This repository contains research and tools for working with Neural Radiance Fields (NeRF) using current methods.

Setup

1. Clone the repository (including submodules):

   git clone --recursive https://github.com/patriklabs/nerf-research.git

Generate Initial Sparse Reconstruction

1. Build the COLMAP Docker image:

   cd colmap && ./build_docker.sh

2. Create a sparse reconstruction of a dataset:

   Run COLMAP on your data to generate a sparse reconstruction by specifying the path to your dataset:

   cd colmap && ./docker_run.sh /path/to/data

Running NeRF with Docker

1. Prepare Configuration:

   • Add or update a configuration file in the config folder to specify your training parameters.

2. Build and Run the Docker Image:

   • Build the Docker image:

     ./docker/build.sh

   • Start a training session:

     ./docker/run.sh /path/to/dataset /path/to/config.yaml

   • Optional: To export a 3D mesh, specify a checkpoint file and add the --visualize flag:

     ./docker/run.sh /path/to/dataset /path/to/config.yaml /path/to/ckpt --visualize

   Example:

   ./docker/run.sh /database config/nerf_config.yaml

Running NeRF with Visual Studio Code (VSCode)

1. Set Up VSCode and Dev Containers:

   • Download and install VSCode if not already installed.
   • Install the Dev Containers extension in VSCode.

2. Configure DevContainer:

   • Update devcontainer.json to map your local data directory to the /database folder inside the container:

     "runArgs": [
        "-v=/path/to/database:/database:rw",
     

3. Run Training and Visualization Tasks:

   • Update or add a configuration file in the config folder.
   • Launch the NeRF Training task to start training.
   • To export a 3D mesh, use the NeRF Visualization task.