gStatLight

A command-line tool to spotlight central objects in 3D Gaussian Splatting (3DGS) reconstructions. Given a walk-around video, gStatLight isolates the subject (e.g., a car) by keeping only Gaussians within a small radius of the scene’s origin.

💾 Installation

1. Create & activate the Conda environment

conda create -n gsl python
conda activate gsl

2. Install dependencies

# Change directories to project root (gStatLight/):
cd <project-dir: gStatLight>

# Install the gStatLight package and its CLI entrypoints:
pip install .

📂 File Structure (Input Layout)

The tool requires the following structure:

gCull/
├── data/
│   └── <experiment-name>/
│       ├── colmap/
│       ├── images/
│       └──  transforms.json
|
└── outputs/
    └── <experiment-name>/
        └── splatfacto/
            └── <model-name>/
                └── config.yml          ← point to this config file for `statlight`
                └── {...}_statlight.ply ← statlight output saved here

🚀 Execution

From your project root:

gsl statlight \
  --load-config <path/to/config.yml>

The final culled 3DGS model is saved alongside your config.yml as {experiment_name}_{model_name}_statlight.ply file.

🛠️ Acknowledgements

This work is built upon and heavily modifies the Nerfstudio/Splatfacto codebases.