By submitting, you confirm that you have read and agree to the sunvage-vertens3d-simhome Terms of Use: the data is for non-commercial research only; redistribution requires prior written permission; results based on the data must cite the source per the README citation format. You will also be asked to share your Hugging Face username and email. If you cannot use the online form, send a request email to chendebi@3vjia.com as described in the README.

sunvage-vertens3d-simhome

English | 中文

Physically-annotated indoor scene USD assets derived from 14,169 real interior design schemes on the sunvage floorplan platform — every furniture instance carries collision shapes, mass, friction and restitution, and opens directly in USD physics simulators.

⚡

12,169

single-room packages
lightweight training & interaction

🏠

2,000

whole-home scenes
full multi-space floorplans

🧱

Physics

per-instance rigid bodies
collision · mass · friction

🗺️

provincial regions
real schemes across China

Dataset description

sunvage-vertens3d-simhome is a collection of physically-annotated indoor scene USD assets derived from real interior design schemes in the sunvage (三维家) floorplan library. It provides both whole-home multi-space scenes and individual single-space scenes. Every physical instance carries collision shapes and mass properties, and the scenes can be loaded directly into USD physics simulators such as Isaac Sim / Isaac Lab for embodied AI, robot simulation, scene understanding, and layout generation research.

Data source

All floorplan schemes come from the sunvage floorplan library — real interior design schemes on the sunvage platform (including structured design data such as plumbing, doors/windows, and cabinetry).
Furniture and fixtures are instantiated as standalone USD assets with collision definitions (collision.json) and rigid-body physical properties (physical.json).
Every instance carries mass, density, center of mass, static/dynamic friction, and restitution coefficients, each annotated with its provenance (estimated / derived / policy) and confidence score. Mass and center of mass are computed from collision geometry (density assigned per furniture category and material); friction comes from a material-classification model with per-instance confidence and a needs_review flag. Structural elements (walls, floors) follow a static-body policy.
Access is granted for research purposes only upon application (see Access Request).

Dataset structure

Scenes are organized by scene ID (scene_*). Each scene directory contains two types of archives:

<scene_id>_single_room_<room-hash>.zip   # Single-space scene package
<scene_id>_multi_space.zip               # Multi-space scene package

After extraction, each archive yields a self-contained top-level directory named after the archive:

<scene_id>_single_room_<room-hash>/
├── scene.usd                     # Ready-to-open USD scene
├── scene.manifest.json           # Coordinate system, space info, export statistics
└── assets/<instance_id>/         # One directory per physical instance
    ├── asset.usd / model.usd     # Instance USD and model USD
    ├── models/model.glb          # Original mesh (converted to USD on first open)
    ├── collision.json            # Collision shape definitions
    └── physical.json             # Rigid-body type, mass policy, etc.

<scene_id>_multi_space/
├── scene.usd
├── scene.manifest.json           # Includes full floorplan (space boundaries, walls, doors/windows)
└── assets/...

single_room vs. multi_space

Directory	Content	Typical use
`single_room`	An independent space (possibly including connected balconies etc.), containing only instances within that space	Single-space navigation, interaction, furniture layout generation, lightweight training
`multi_space`	Whole-home complete scene USD with all spaces and cross-space structure	Multi-space navigation, whole-home tasks, scene-level understanding

Additional notes:

scene.manifest.json records the coordinate system: right-handed, Z-up, XY horizontal plane, units in meters (consistent with USD defaults).
Spaces use functional names (e.g., 卧室/bedroom, 卫生间/bathroom, 客餐厅/living-dining room, 厨房/kitchen), with hash suffixes to distinguish same-named spaces.

Coverage

Single-space coverage (single_room packages)

Space type	Packages	Share	Median area	Area range
Bedroom	5,048	41.5%	10.9 m²	4.0–75.5 m²
Bathroom	3,133	25.7%	3.9 m²	0.9–29.2 m²
Living-dining room	1,997	16.4%	36.3 m²	2.9–123.4 m²
Kitchen	1,677	13.8%	5.7 m²	1.8–38.5 m²
Study	296	2.4%	7.6 m²	3.4–41.3 m²
Others (storage, balcony, etc. as standalone packages)	18	0.1%	—	—
Total	12,169	100%	—	—

Area distribution of single-space packages:

Package area	Packages	Share
< 5 m²	3,154	25.9%
5–10 m²	3,770	31.0%
10–20 m²	3,021	24.8%
20–40 m²	1,455	12.0%
≥ 40 m²	769	6.3%

Package areas range 0.1–123.4 m², median 8.7 m²; the largest packages are living-dining rooms (including merged corridors etc.).
Each home yields 2–12 single-space packages, typically 5–8.

Multi-space coverage (multi_space packages)

Floor area	Homes	Share
< 60 m²	164	8.2%
60–80 m²	427	21.4%
80–100 m²	578	28.9%
100–120 m²	428	21.4%
120–150 m²	247	12.4%
≥ 150 m²	156	7.8%
Total	2,000	100%

Areas range 33.1–277.3 m², with the median band at 80–100 m²; intervals are left-closed, right-open (e.g., 60–80 means 60 ≤ area < 80).
Bedroom counts cover 1–5 rooms: 1-room 274, 2-room 665, 3-room 628, 4-room 311, 5-room 122.
Raw space-type distribution (for reference, including auxiliary spaces merged into primary spaces): bedrooms 5,048, bathrooms 3,133, balconies 2,012, living-dining rooms 1,997, kitchens 1,677, studies 296, storage rooms 222, entry gardens 106, walk-in closets 62, corridors 23.
Homes come from 31 provincial-level regions across China.

Usage

# Single-room packages (requires granted access)
hf download sunvage-lab/sunvage-vertens3d-simhome --repo-type dataset \
  --include "single_room/*" --local-dir simhome

# Whole-home multi-space scenes
hf download sunvage-lab/sunvage-vertens3d-simhome --repo-type dataset \
  --include "multi_space/*" --local-dir simhome

Access Request

Online request (recommended): click the "Agree to terms and submit request" button at the top of this page. The form collects: name, purpose (individual/company), company name and job title (required for company requests), terms confirmation, plus your Hugging Face username and email.
Email request (fallback): if you cannot use the online form, follow the template below and send a request email to chendebi@3vjia.com.
Once approved, your Hugging Face account will be added to the authorized list and can download the data; if declined, you will see a rejection notice.

Email request format (plain text is fine):

Subject: [sunvage Scene Physics Data Access Request] <Organization/Name>

Name:
Organization / Team:
Hugging Face username:
Email:
Position / role (student, researcher, engineer, etc.):

Intended use (research/evaluation/product testing, 2-3 sentences):
Application scenario (navigation / simulation / layout generation / other):
Will you train models published externally on this data: Yes / No
Do you need a commercial-use evaluation: Yes / No

Applicant commitments:
1. The data will be used only for the declared purpose and not redistributed to unauthorized third parties;
2. Results involving this dataset will cite the data source per the README citation format;
3. You acknowledge and agree to comply with the dataset Terms of Use.

Terms of Use (License)

This dataset is released under the sunvage-vertens3d-simhome Terms of Use (custom terms, license: other). The full text is distributed in the repository's License.md.

Citation

If this dataset helps your research, please cite:

@misc{sunvage-vertens3d-simhome-2026,
  title        = {sunvage-vertens3d-simhome: A Physically-Annotated Indoor USD Scene Dataset},
  author       = {sunvage Research},
  year         = {2026},
  howpublished = {\url{https://huggingface.co/datasets/sunvage-lab/sunvage-vertens3d-simhome}}
}

Contact

Data questions and access requests: chendebi@3vjia.com

中文

sunvage-vertens3d-simhome 是一组带物理信息的室内场景 USD 资产库，源自三维家（sunvage）户型库中的真实户型设计方案：既提供整户多空间场景，也提供单空间场景；每个物理实例附带碰撞体与质量等物理属性，可直接加载到 Isaac Sim / Lab 等 USD 物理仿真器中，用于具身智能、机器人仿真、场景理解与布局生成研究。

数据来源

所有户型方案均来自三维家户型库，是三维家平台上的真实室内设计户型方案（含水路、门窗、柜体等结构化设计数据）。
户型物体：实例化为独立 USD 资产，并存在碰撞体（collision.json）与刚体物理属性（physical.json）。
每个实例均附带质量、密度、质心、静/动摩擦系数与恢复系数，且逐项标注数值来源（估算/推导/策略）与置信度。质量与质心由碰撞体几何计算（密度按家具类别与材质赋值）；摩擦系数由材质分类模型给出，带逐实例置信度与 needs_review 标记。墙体、地面等结构件按 static 刚体策略处理。
数据仅限研究用途申请使用（见下方数据申请）。

数据集结构

数据集按场景 ID（scene_*）组织，每个场景目录下含两类压缩包：

<scene_id>_single_room_<room-hash>.zip   # Single-space scene package
<scene_id>_multi_space.zip               # Multi-space scene package

每个压缩包解压后是一个与包同名的自包含顶层目录：

<scene_id>_single_room_<room-hash>/
├── scene.usd                     # Ready-to-open USD scene
├── scene.manifest.json           # Coordinate system, space info, export statistics
└── assets/<instance_id>/         # One directory per physical instance
    ├── asset.usd / model.usd     # Instance USD and model USD
    ├── models/model.glb          # Original mesh (converted to USD on first open)
    ├── collision.json            # Collision shape definitions
    └── physical.json             # Rigid-body type, mass policy, etc.

<scene_id>_multi_space/
├── scene.usd
├── scene.manifest.json           # Includes full floorplan (space boundaries, walls, doors/windows)
└── assets/...

single_room 与 multi_space 的区别

目录	内容	适用场景
`single_room`（单空间）	场景是一个独立的空间（可能包含联通的阳台等），仅含该空间内的实例	单空间导航、交互、家具布局生成、轻量训练
`multi_space`（多空间）	整户完整场景 USD，含全部空间与跨空间结构	多空间导航、全屋任务、场景级理解

补充说明：

scene.manifest.json 中记录坐标系：右手系、Z 轴向上、XY 水平面、单位米（与 USD 默认一致）。
空间名使用功能命名（如 卧室、卫生间、客餐厅、厨房），并以哈希后缀区分同名空间。

覆盖范围

单空间覆盖（single_room 包）

共 12,169 个单空间包：

主空间类型	单空间包数	占比	面积中位数	面积范围
卧室	5,048	41.5%	10.9 m²	4.0–75.5 m²
卫生间	3,133	25.7%	3.9 m²	0.9–29.2 m²
客餐厅	1,997	16.4%	36.3 m²	2.9–123.4 m²
厨房	1,677	13.8%	5.7 m²	1.8–38.5 m²
书房	296	2.4%	7.6 m²	3.4–41.3 m²
其他（储物间、阳台等独立成包）	18	0.1%	—	—
合计	12,169	100%	—	—

单空间包的面积分布：

包面积	包数	占比
< 5 m²	3,154	25.9%
5–10 m²	3,770	31.0%
10–20 m²	3,021	24.8%
20–40 m²	1,455	12.0%
≥ 40 m²	769	6.3%

包面积 0.1–123.4 m²，中位数 8.7 m²；面积最大的是客餐厅包（含并入的过道等）。
每个户型产出 2–12 个单空间包，集中在 5–8 个。

多空间覆盖（multi_space 包）

共 2000 个多空间包：

套内面积	户型数	占比
< 60 m²	164	8.2%
60–80 m²	427	21.4%
80–100 m²	578	28.9%
100–120 m²	428	21.4%
120–150 m²	247	12.4%
≥ 150 m²	156	7.8%
合计	2000	100%

面积范围 33.1–277.3 m²，中位区间为 80–100 m²；区间为左闭右开（如 60–80 表示 60 ≤ 面积 < 80）。
卧室数覆盖 1–5 室：1 室 274、2 室 665、3 室 628、4 室 311、5 室 122。
原始空间类型分布（参照，含被并入主空间的附属空间）：卧室 5,048、卫生间 3,133、阳台 2,012、客餐厅 1,997、厨房 1,677、书房 296、储物间 222、入户花园 106、衣帽间 62、过道 23。
户型来自全国 31 个省级行政区。

数据申请

在线申请（推荐）：直接在本页顶部填写申请表单并点击"同意条款并提交申请"。表单会收集：姓名、申请用途（个人/公司）、公司名与职位（公司申请需填）、条款确认，以及你的 Hugging Face 用户名与邮箱。
邮件申请（备选）：无法在线填写时，参考本节说明、按下方模板发送申请邮件至 chendebi@3vjia.com。
审核通过后，你的 Hugging Face 账号即被加入授权名单，可下载数据；被拒绝时会看到拒绝提示。

申请邮件请使用以下格式（纯文本即可）：

主题：[三维家场景物理数据使用申请] <机构/姓名>

姓名：
所属机构 / 团队：
Hugging Face 用户名：
邮箱：
职位 / 身份（学生、研究员、工程师等）：

预期用途（研究/评估/产品测试，简述 2-3 句）：
使用场景（导航 / 仿真 / 布局生成 / 其他）：
是否会将数据用于训练对外发布的模型：是 / 否
是否需要商业使用评估：是 / 否

申请人承诺：
1. 数据仅用于上述申报用途，不向未获授权的第三方再分发；
2. 涉及本数据集的成果将按 README 引用格式注明数据来源；
3. 已经知晓并遵守本数据集使用条款。

使用条款（License）

本数据集按 sunvage-vertens3d-simhome 数据使用条款（自定义条款，license: other）发布，全文随仓库 License.md 分发。

引用

如果本数据集对你的研究有帮助，请引用：

@misc{sunvage-vertens3d-simhome-2026,
  title        = {sunvage-vertens3d-simhome: A Physically-Annotated Indoor USD Scene Dataset},
  author       = {sunvage Research},
  year         = {2026},
  howpublished = {\url{https://huggingface.co/datasets/sunvage-lab/sunvage-vertens3d-simhome}}
}

联系方式

数据相关问题与使用申请：chendebi@3vjia.com

Downloads last month: 544

Total file size:

992 GB