shaygeko/spaghetti-splatting-viewer

Spaghetti Splatting viewer model: GeomHair 380

This repository holds one model for the Spaghetti Splatting WebGPU viewer (project page).

• geomhair_380/: hair and bald head of GeomHair subject 380, as Gaussian segments with SV8 colour. The values are stored as float16. The viewer converts them to float32 at load.
• models.json: the list of models that the viewer loads.

Data source

We trained the model on our own Blender renders of the hair strands of subject 380 of Strands400, the strand dataset of GeomHair. Strands400 contains strands that GeomHair reconstructed from NPHM head scans. The bald head is the generic FLAME-topology head prior of GeomHair, not a scan of the subject.

If you use this model, please also cite GeomHair and NPHM:

@misc{lazuardi2026geomhairreconstructionhairstrands,
  title         = {GeomHair: Reconstruction of Hair Strands from Colorless 3D Scans},
  author        = {Rachmadio Noval Lazuardi and Artem Sevastopolsky and Egor Zakharov and Matthias Niessner and Vanessa Sklyarova},
  year          = {2026},
  eprint        = {2505.05376},
  archivePrefix = {arXiv},
  primaryClass  = {cs.CV},
  url           = {https://arxiv.org/abs/2505.05376}
}

@inproceedings{giebenhain2023nphm,
  title     = {Learning Neural Parametric Head Models},
  author    = {Giebenhain, Simon and Kirschstein, Tobias and Georgopoulos, Markos and R{\"u}nz, Martin and Agapito, Lourdes and Nie{\ss}ner, Matthias},
  booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
  year      = {2023}
}

Scenes: Mip-NeRF 360 room

• scenes/models.json: the list of scene models that the scene viewer loads.
• scenes/room_300k/: the room scene of Mip-NeRF 360, trained with a cap of 300k segments.
• scenes/room_full/: the room scene of Mip-NeRF 360, trained with the full segment cap (1.5M).

Both models have Gaussian segments with SV8 colour and the per-segment opacity for the clay view. The SV8 values are stored as float16.

If you use these models, please also cite Mip-NeRF 360:

@inproceedings{barron2022mipnerf360,
  title     = {Mip-NeRF 360: Unbounded Anti-Aliased Neural Radiance Fields},
  author    = {Barron, Jonathan T. and Mildenhall, Ben and Verbin, Dor and Srinivasan, Pratul P. and Hedman, Peter},
  booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
  year      = {2022}
}

Scenes: sauce

• scenes/sauce/: our sauce capture (spaghetti plate, mug and kettle), Beta kernel, SV8 colour (float16). radius.f32 holds the clay tube radius of each segment: the spaghetti tubes have the half-maximum radius of their kernel; every other segment has 0.5 mm.