SAGE-3D InteriorGS USDZ: USDZ-Format 3D Gaussian Scenes for Isaac Sim

Paper | Project Page | Code

InteriorGS dataset converted to USDZ format for seamless integration with NVIDIA Omniverse and Isaac Sim platforms.

---

📢 News

• 2025-12-15: Released SAGE-3D InteriorGS USDZ dataset with 1,000 converted scenes.

---

📋 Overview

While the original InteriorGS dataset provides high-quality 3D Gaussian Splatting scenes in compressed PLY format, these files are not directly compatible with modern simulation platforms like NVIDIA Isaac Sim and Omniverse. To bridge this gap, we present SAGE-3D InteriorGS USDZ, a format-converted version of the entire InteriorGS dataset.

This dataset provides:

• 1,000 indoor scenes in USDZ format ready for Isaac Sim 5.0+
• Photorealistic rendering quality preserved from original 3DGS data
• Direct compatibility with NVIDIA Omniverse and Isaac Sim

Conversion Pipeline

The conversion is performed using NVIDIA's 3DGRUT library:

InteriorGS compressed PLY → Decompressed PLY → USDZ (3DGRUT)

The USDZ format uses an extension of the UsdVolVolume Schema specifically designed for 3D Gaussian rendering in Isaac Sim, enabling:

Real-time rendering - Leverage Isaac Sim's optimized 3DGS renderer
Physics simulation - Combine with collision meshes for embodied AI
Platform compatibility - Work with Omniverse ecosystem tools

---

🗂️ Dataset Structure

InteriorGS_usdz/
├── 839873.usdz        # Scene in USDZ format
├── 839874.usdz
├── 839875.usdz
└── ...                # 1,000 scenes total

---

✨ Sample Usage

This section demonstrates how to use this USDZ dataset within the SAGE-3D framework, specifically how to build USDA scene files with integrated collision bodies for Isaac Sim.

Prerequisites:

1. Install splat-transform and 3DGRUT as described in the project's GitHub repository.
2. Download the original InteriorGS dataset (compressed PLY files) and the SAGE-3D Collision Mesh Dataset.

Step 1: Convert Compressed PLY to Original PLY (if starting from original InteriorGS)

# Example: Convert a single scene
splat-transform /path/to/InteriorGS/0001_839920/3dgs_compressed.ply \
    /path/to/output/0001_839920.ply

Step 2: Convert PLY to USDZ (if starting from original InteriorGS)

# Example: Convert a single scene
python -m threedgrut.export.scripts.ply_to_usd \
    /path/to/ply/0001_839920.ply \
    --output_file /path/to/usdz/0001_839920.usdz

Note: This dataset (spatialverse/SAGE-3D_InteriorGS_usdz) already provides the USDZ files, so you can directly use them instead of performing Steps 1 and 2.

Step 3: Build USDA Scene Files with Collision Meshes

To use the USDZ scenes with physics simulation in Isaac Sim, they need to be combined with collision meshes and converted to USDA format.

python Code/benchmark/scene_data/sage3d_usda_builder.py \
    --usdz-dir /path/to/usdz_dataset \
    --out-dir /path/to/output/usda \
    --template Data/template.usda \
    --usdz-placeholder "@usdz_root[gauss.usda]@" \
    --collision-placeholder "@collision_root@" \
    --usdz-path-template "/path/to/usdz_dataset/{scene_id}.usdz[gauss.usda]" \
    --collision-path-template "/path/to/collision_mesh_dataset/{scene_id}/{scene_id}_collision.usd" \
    --overwrite

Replace /path/to/usdz_dataset with the local path to this Hugging Face dataset, and /path/to/collision_mesh_dataset with the local path to spatialverse/SAGE-3D_Collision_Mesh.

---

🔗 Related Datasets

This dataset is part of the SAGE-3D project:

1. InteriorGS: Original 3DGS scenes with semantic annotations
   → spatialverse/InteriorGS

2. SAGE-3D InteriorGS USDZ (This dataset): USDZ format for Isaac Sim
   → spatialverse/SAGE-3D_InteriorGS_usdz

3. SAGE-3D Collision Mesh: Physics-enabled collision bodies
   → spatialverse/SAGE-3D_Collision_Mesh

4. SAGE-3D VLN Data: Navigation trajectories and instructions
   → spatialverse/SAGE-3D_VLN_Data

---

📄 License

This dataset is released under CC-BY-NC-4.0.

---

🤝 Acknowledgments

Format conversion was performed using NVIDIA's 3DGRUT library. We thank the NVIDIA Toronto AI Lab for developing and open-sourcing this excellent tool.

---

📜 Citation

If you use SAGE-3D InteriorGS USDZ in your research, please cite:

Our Paper:

@inproceedings{miao2026towards,
  title = {Towards Physically Executable {3D} Gaussian for Embodied Navigation},
  author = {Miao, Bingchen and Wei, Rong and Ge, Zhiqi and Sun, Xiaoquan and Gao, Shiqi and Zhu, Jingzhe and Wang, Renhan and Tang, Siliang and Xiao, Jun and Tang, Rui and Li, Juncheng},
  booktitle = {The Fourteenth International Conference on Learning Representations},
  year = {2026},
  url = {https://openreview.net/forum?id=HB6KvsqcAn}
}

Please also cite the InteriorGS dataset:

@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}}
}

---