livin-benchmark/LIVIN-Scenes

Dataset Overview

This release contains indoor scene assets from LIVIN, including assembled Blender scenes, room structures, object meshes, object metadata, poses, and product reference images.

Content	Count
Scenes	30
Rooms	319
Objects	13,000+
Objects labeled articulated	1,500+

File Structure

LIVIN_Scenes/
├── README.md
├── LICENSE
├── source_blend/
│   ├── scene_001.blend
│   ├── ...
│   └── scene_030.blend
└── scenes/
    └── scene_XXX/
        ├── labels.json
        ├── products/
        │   └── <image_filename>.png
        └── <room_name>/
            ├── pose.json
            └── GLB/
                ├── room_structure.glb
                └── obj_sXXX_NNNN.glb

Each scene has one assembled .blend file and one corresponding directory under scenes/.

File or directory	Purpose
source_blend/scene_XXX.blend	The assembled scene, organized under the top-level collections structure, articulated, and non_articulated. Architectural geometry is merged into one scene-level structure mesh.
scenes/scene_XXX/labels.json	Scene-wide metadata, with separate objects and structures arrays.
scenes/scene_XXX/<room_name>/GLB/room_structure.glb	Architectural geometry for that room.
scenes/scene_XXX/<room_name>/GLB/obj_sXXX_NNNN.glb	An individual object asset, including its internal node hierarchy and animation where present.
scenes/scene_XXX/<room_name>/pose.json	Object and room-structure placement matrices, keyed by ID.
scenes/scene_XXX/products/	Scene-level product reference images shared across rooms and linked from labels.json.

Field Definitions

Each record in the objects or structures array of labels.json has these fields:

Field	Meaning
id	Object ID in this release, or the room-local ID room_structure.
category	English semantic category.
name	English description of the specific instance.
mesh	GLB path relative to the scene directory.
product_image	Product reference image path relative to the scene directory.
motion_type	articulated or non_articulated for an asset.

Example object record:

{
  "id": "obj_s002_0066",
  "category": "pump_bottle",
  "name": "hand soap in a white square bottle with a blue pump",
  "mesh": "washroom/GLB/obj_s002_0066.glb",
  "product_image": "products/s010_w_bluepump.png",
  "motion_type": "non_articulated"
}

pose.json contains only ID-to-matrix mappings. Object descriptions and file references are stored in labels.json.

Citation

If you use LIVIN in your research, please cite LIVIN: Benchmarking Spatial and Embodied Intelligence in Digital Twins of Lived-In Homes.

@misc{xu2026livinbenchmarkingspatialembodied,
      title={LIVIN: Benchmarking Spatial and Embodied Intelligence in Digital Twins of Lived-In Homes}, 
      author={Peijun Xu and Chuansen Nie and Yiyang He and Yinuo Bai and Jingyang Liu and Kuixiang Shao and Yuyang Jiao and Kuanhao Xia and Jiayi Zhu and Zitian Yang and Yanqi Zhang and Tianye Tan and Shuwei Di and Junyi Xu and Jingyi Yu and Jiayuan Gu},
      year={2026},
      eprint={2610.12069},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2610.12069}, 
}