Development-stage release / 开发阶段发布。 This Hugging Face repository is currently used for ObjGauss development, verification, and reproducibility handoff. Assets, file layout, metrics, model weights, and documentation may change before a stable release.

ObjGauss NeRF Lego near-1M Splatfacto Checkpoint

This model repository contains the Nerfstudio Splatfacto checkpoint and config used to export the ObjGauss near-1M NeRF Lego Gaussian asset.

The downloadable object-aware Gaussian assets, Object Field files, masks, manifests, and metrics are published in the companion dataset repository:

jianyong365/objgauss-nerf-lego-near1m

Contents

  • nerfstudio/config.yml: Nerfstudio Splatfacto training config.
  • nerfstudio/dataparser_transforms.json: dataparser transform metadata.
  • nerfstudio/nerfstudio_models/step-000009999.ckpt: final training checkpoint.
  • checksums.sha256: SHA256 checksums for uploaded model files.

TensorBoard event files are intentionally omitted from this release. The exported .splat, object-aware PLY, Object Field .npz, masks, manifests, and metrics live in the companion dataset repository.

Training Summary

  • Scene: NeRF Synthetic Lego
  • Trainer: Nerfstudio Splatfacto
  • Iterations: 10,000
  • Exported Gaussian count: 4,503,634
  • Tuned initialization: random-init with num_random=1,300,000
  • ObjGauss Object Field slots: 4

Limitations

This is a research artifact derived from NeRF Synthetic Lego. Downstream use must respect the upstream data terms. The companion dataset contains current ObjGauss metrics and WebGPU C-path production SLA status; the SLA is not marked as passed in this release.

Source Code

https://github.com/sessgraph/ObjGauss
Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Dataset used to train jianyong365/objgauss-nerf-lego-near1m-model