---
license: apache-2.0
language:
- zh
pipeline_tag: text-to-3d
---


# 🪑 3D Furniture Generation Model | 3D 家具生成模型

![License](https://img.shields.io/badge/license-Apache--2.0-green)
![HuggingFace](https://img.shields.io/badge/HuggingFace-Model-yellow)
![PyTorch](https://img.shields.io/badge/PyTorch-2.0+-red)

---

## 📌 项目简介 | Introduction
**中文**  
本项目是一个基于 Transformer 架构的 **3D 家具生成大模型**，支持根据输入关键词（如 `衣柜`、`橱柜`）生成对应的家具 3D 结构数据。  
该模型可直接生成自定义 JSON 结构，包含家具的 **面板尺寸、位置、旋转、类型** 等信息，可在 Three.js 等前端框架中直接渲染。  
本模型已在 Hugging Face Model Hub 开源，欢迎下载、测试和二次训练。

**English**  
This project is a **3D furniture generation model** based on a Transformer architecture.  
It supports generating corresponding 3D furniture structural data based on text prompts such as `"wardrobe"` or `"cabinet"`.  
The model directly produces custom JSON structures containing **panel size, position, rotation, and type** information, which can be rendered directly in front-end frameworks such as Three.js.  
The model is open-sourced on Hugging Face Model Hub, and you are welcome to download, test, and fine-tune it.

---

## 🚀 功能特点 | Features
**中文**
- **结构化输出**：生成可直接解析的 JSON 结构
- **参数化控制**：支持输入家具类型、尺寸等参数
- **高精度坐标**：基于归一化训练，精度可达毫米级
- **可扩展性强**：可基于自有家具数据进行增量训练
- **适配 Three.js**：可直接在浏览器中渲染生成的家具模型

**English**
- **Structured Output** – Generates JSON structures that can be directly parsed.
- **Parametric Control** – Supports specifying furniture type, size, and other parameters.
- **High-Precision Coordinates** – Trained with normalization to achieve millimeter-level accuracy.
- **Highly Extensible** – Can be fine-tuned with your own furniture dataset.
- **Three.js Compatible** – The generated models can be rendered directly in the browser.

---

## 📂 数据示例 | Example Data
```json
{
  "Name": "三门衣柜 / Three-door Wardrobe",
  "Panels": [
    {
      "Type": "左侧板 / Left Side Panel",
      "Size": [0.018, 2.000, 0.600],
      "Position": [0.000, 1.000, 0.300],
      "Rotation": [0.0, 0.0, 0.0]
    }
  ]
}
```

---

## 📧 联系方式 | Contact

### 中文
如需合作或技术交流，请联系邮箱：**chengyuhan1030@gmail.com**

### English
For collaboration or technical inquiries, please contact: **chengyuhan1030@gmail.com**