splatter-server

Splatter compute node for the Auki Network: runs Gaussian splatting jobs (COLMAP + nerfstudio/splatfacto) as part of the reconstruction pipeline.

License

This project is licensed under the MIT License.

Contributing

See CONTRIBUTING.md for how to report issues, open PRs, and run the Rust checks locally.

Docker

Build

From repo root path:

docker build -t splatter-server .

Run

Create a .env from server/rust/.env.example and set your DMS/DDS URLs and registration secret, then:

docker run --gpus all -p 8080:8080 --name splatter -d --env-file .env splatter-server

Run Trainer

python3 run.py \
--domain_id {domain_id}
--job_id {job_id} \
--job_root_path {path/to/job/root} \
--log_level {log level}

Required Files

# Input Files
{job_root_path}
├── datasets
│   └── {dataset}
│       └── Frames.mp4
├── refined
│   └── global
│       └── refined_sfm_combined
│           ├── cameras.bin
│           ├── images.bin
│           └── points3D.bin

Output Files

# Output Files
{job_root_path}
├── Frames
│   ├── {images}
│   └── ...
├── refined
│   ├── nerfstudio-data
│   │   └── {converted nerfstudio data from colmap}
│   └── splatter
│       ├── splat.ply
│       ├── splat_rot.ply
│       ├── splat_rot.splat # this is what needs to be uploaded to dmt
│       └── splatfacto
│           └── {splat torch model}