---
license: other
license_name: interiorgs-terms-of-use
license_link: https://kloudsim-usa-cos.kujiale.com/InteriorGS/InteriorGS_Terms_of_Use.pdf
task_categories:
  - other
language:
  - en
tags:
  - 3dgs
  - indoor-scene
  - scene-understanding
pretty_name: InteriorGS Preprocessed
extra_gated_prompt: |
  This repository contains processed data derived from the original InteriorGS release.

  Before accessing this repository, please confirm that you will comply with the original InteriorGS Terms of Use.

  If you use this processed benchmark in the Chorus setting, please also cite the Chorus paper.
extra_gated_fields:
  Full name: text
  Institutional email: text
  "I agree to the original InteriorGS Terms of Use": checkbox
---

# InteriorGS Preprocessed

This repository provides processed InteriorGS data used for additional evaluation in **Chorus**.

The data is derived from the original [InteriorGS](https://huggingface.co/datasets/spatialverse/InteriorGS) release. We convert the released 3DGS scenes into per-scene `*.npy` files following the format used in the SceneSplat/Chorus codebase.

For each scene, we process the original 3D bounding-box annotations and assign:

- `segment.npy`: semantic labels for Gaussian rows
- `instance.npy`: instance labels for Gaussian rows

Labels are 0-indexed, and the ignore label is `-1`. The label assignment uses connected components to improve spatial consistency.

For semantic evaluation, we map the original InteriorGS semantic taxonomy into a 72-class benchmark taxonomy. The mapping file is provided at:

```text
metadata/semantic_mapping.csv
````

The 72 class names are provided at:

```text
metadata/taxonomy_labels.txt
```

The row index in `taxonomy_labels.txt` corresponds to the class index used in `segment.npy`.

We use the InteriorGS `test` split as the benchmark split in the Chorus paper. Split files are provided at:

```text
metadata/splits
```

## Terms of Use

This processed dataset is derived from InteriorGS and follows the original InteriorGS Terms of Use. Please use it only for non-commercial research and educational purposes, do not redistribute the downloaded data, and cite the relevant works.

## Citation

If you use this processed data, please consider citing InteriorGS and Chorus.

```bibtex

@misc{InteriorGS2025,
  title        = {InteriorGS: A 3D Gaussian Splatting Dataset of Semantically Labeled Indoor Scenes},
  author       = {SpatialVerse Research Team, Manycore Tech Inc.},
  year         = {2025},
  howpublished = {\url{https://huggingface.co/datasets/spatialverse/InteriorGS}}
}

@InProceedings{Li_2026_CVPR,
    author    = {Li, Yue and Ma, Qi and Yang, Runyi and Ma, Mengjiao and Ren, Bin and Popovic, Nikola and Sebe, Nicu and Gevers, Theo and Van Gool, Luc and Paudel, Danda Pani and Oswald, Martin R.},
    title     = {Chorus: Multi-Teacher Pretraining for Holistic 3D Gaussian Scene Encoding},
    booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
    month     = {June},
    year      = {2026},
    pages     = {21431-21442}
}
```