SAGE-10k

# SAGE-10k <p> <strong>SAGE-10k</strong> is a large-scale interactive indoor scene dataset featuring realistic layouts, generated by the agentic-driven pipeline introduced in "SAGE: Scalable Agentic 3D Scene Generation for Embodied AI". The dataset contains <strong>10,000 diverse scenes</strong> spanning 50 room types and styles, along with <strong>565K uniquely generated 3D objects</strong>. </p> <img src="assets/preview_and_stats_v2.png" alt="Teaser" width=100% > <img src="assets/preview_grid.jpg" alt="Teaser" width=100% > ## 🔑 Key Features <div class="section"> <p>SAGE-10k integrates a wide variety of scenes, and particularly, preserves small items for complex layouts, resolves collisions, and further incorporates interactive objects, thus ensures:</p> <ul> <li>📊 <strong>Large scale</strong>: 10,000 diverse scenes including 565K generated 3D objects by TRELLIS.</li> <li>🏠 <strong>Diverse room types and styles</strong>: Covers 50 different room types and interior design styles for broad scene variability.</li> <li>🔧 <strong>Simulation-ready</strong>: Scenes are structured for direct use in physics-based simulators and embodied AI environments.</li> </ul> </div> ### Supported Tasks <ul> <li> ✅ 3D scene reconstruction <li> ✅ 3D scene understanding <li> ✅ Embodied navigation & manipulation <li> ✅ Vision-language grounding </ul> ## 📋 Table of Contents - [📊 Dataset Statistics](#-dataset-statistics) - [🔑 Key Features](#-key-features) - [⚙️ Getting Started](#-getting-started) - [Download the Dataset](#download-the-dataset) - [Dataset Structure](#dataset-structure) - [🧷 Citation](#-citation) - [📄 License](#-license) ## ⚙️ Getting Started ### Download the Dataset To download the full dataset, you can use the following code. If you encounter any issues, please refer to the official Hugging Face documentation. ``` huggingface-cli login --token YOUR_HF_TOKEN_HERE huggingface-cli download nvidia/SAGE-10k --repo-type dataset --local-dir SAVE_DIR ``` ### Dataset Structure ```shell SAGE-10k/ |-- scenes/ # All the generated scenes |-- scene_id/ # Each generated scene |-- objects/ # Geometry and texture of each generated object |-- materials/ # Textures of room walls, door, and floor. |-- preview/ # Preview visualization of the room. |-- layout_id.json/ # Information of the generated scene. ``` ### Quick Start You can use kits in kits/examples.sh to generate glb, usd files, as well as render video with the generated camera trajectory and load into IsaacSim. ### Notes This dataset is purely agentic-driven generated from SAGE without any manual filtering. The quality of every scene might be varied. ## 🧷 Citation ```BibTex @article{xia2026sage, title={SAGE: Scalable Agentic 3D Scene Generation for Embodied AI}, author={Xia, Hongchi and Li, Xuan and Li, Zhaoshuo and Ma, Qianli and Xu, Jiashu and Liu, Ming-Yu and Cui, Yin and Lin, Tsung-Yi and Ma, Wei-Chiu and Wang, Shenlong and Song, Shuran and Wei, Fangyin}, journal={arXiv preprint arXiv:2602.10116}, year={2026} } ``` ## 📄 License This dataset is released under the [Apache License 2.0](https://www.apache.org/licenses/LICENSE-2.0). You are free to use, modify, and distribute this dataset for both commercial and non-commercial purposes, provided that proper attribution is given.