🧵 3DFDReal: Real-World 3D Fashion Dataset

Empowering Virtual Try-On Applications with High-Quality 3D Fashion Data

Electronics and Telecommunications Research Institute (ETRI)
Media Intellectualization Research Section

---

🌟 Highlights

📊 1,000+ 3D Point Clouds High-quality captures

🎥 4K@60fps Multi-View Videos Professional recording

🏷️ Rich Metadata Detailed Annotations Semantic labels & attributes

🎮 Metaverse Ready ZEPETO Compatible Direct deployment support

---

🎯 What is 3DFDReal?

3DFDReal is a groundbreaking real-world fashion dataset designed for cutting-edge 3D vision research. Bridging the gap between high-quality 3D fashion modeling and practical virtual environment deployment, our dataset provides researchers and developers with:

• ✨ Individual fashion items and complete outfit combinations
• 🎭 Gender-balanced mannequin representations
• 🔧 Rigging-ready T-pose and upright pose variations
• 📝 Comprehensive metadata including semantic attributes and structured segmentations

Perfect for applications in:

• 🛍️ Virtual Try-On Systems
• 👤 Avatar Modeling & Customization
• 🎨 Digital Fashion Design
• 🌐 Metaverse Asset Creation
• 🤖 Pose-Aware 3D Understanding

---

🔬 Data Collection Pipeline

Pipeline Stages:

1. 📦 Asset Selection
   Curated selection of fashion items with detailed tagging (individual items & complete sets)

2. 📹 Recording Setup
   Professional capture using iPhone 13 Pro with controlled lighting and multi-angle coverage

3. ☁️ 3D Ground Truth Generation
   High-fidelity point cloud generation with manual segmentation using professional 3D tools

4. 🎮 Application & Validation
   Rigging and deployment testing in real metaverse environments (ZEPETO)

---

📈 Dataset Statistics

👔 Fashion Class Distribution

Distribution of fashion items across the dataset, with pants and sweatshirts being the most represented categories

🎭 Combination Analysis


Sneakers and pants appear most frequently in mannequin outfit combinations

📊 Dataset Composition

Key Insights:

• 📦 Average items per outfit: 4 distinct fashion pieces
• ⚖️ Gender balance: Equal representation across combinations
• 🕴️ Pose distribution: Upright poses (standard) + T-poses (rigging-optimized)

---

📂 Dataset Organization

3DFDReal/
│
├── 🎨 Assets (Individual Items)
│   ├── PointCloud_Asset/     # Raw .ply point clouds
│   ├── Video_Asset/          # 3D rotation videos
│   └── Label_Asset/          # Category & class labels
│
├── 👕 Combinations (Full Outfits)
│   ├── PointCloud_Combine/   # Mannequin point clouds
│   │   ├── train/
│   │   ├── val/
│   │   └── test/
│   ├── Video_Combine/        # Mannequin videos
│   └── Label_Combine/        # Combination labels
│
└── 📋 Metadata
    ├── asset_meta.json               # Individual item metadata
    ├── combination_meta.json         # All combinations
    ├── {train,val,test}_combination_meta.json
    └── label_map.csv                # Label mapping reference

🔑 Metadata Schema

Each metadata entry contains:

• label_str: Human-readable class name
• gender: Male/Female/Unisex
• pose: T-pose/Upright
• type: Asset/Combination
• wnlemmas: Fine-grained semantic tags

---

🏆 Benchmark Results

🎯 3D Object Segmentation

Using SAMPart3D as baseline:

Metric	Score
mIoU	0.9930
Average Precision	Class-dependent

🔨 3D Reconstruction

Baseline models evaluated:

• Generation: DDPM-A
• Completion: SVD-SVDFormer

Model	CD	DCD	F1-Score
DDPM-A	0.628±0.887	-	-

---

🚀 Getting Started

Quick Start

# Load the dataset from Hugging Face
from datasets import load_dataset

dataset = load_dataset("kusses/3DFDReal")

# Access different splits
train_data = dataset['train']
val_data = dataset['validation']
test_data = dataset['test']

Example Use Cases

• 🛍️ Virtual Try-On Applications
• 🌐 Metaverse Asset Generation
• 🤖 Pose-Aware Segmentation Research
• 👤 Avatar Rigging & Deformation
• 🎨 Digital Fashion Design Tools

---

📝 Citation

If you use 3DFDReal in your research, please cite:

@misc{3DFDReal2025,
  title={3DFDReal: Real-World 3D Fashion Dataset for Virtual Try-On Applications},
  author={Jiyoun Lim, Jungwoo Son, Alex Lee, Sun-Joong Kim, Nam Kyung Lee, Won-Joo Park},
  year={2025},
  publisher={Hugging Face},
  howpublished={\url{https://huggingface.co/datasets/kusses/3DFDReal}},
  note={Electronics and Telecommunications Research Institute (ETRI)}
}

---

🤝 Contributing

We welcome contributions to improve and expand 3DFDReal! Please feel free to:

• 🐛 Report issues or bugs
• 💡 Suggest new features or improvements
• 🔧 Submit pull requests
• 💬 Join discussions on our Hugging Face page

---

📬 Contact

Jiyoun Lim
Electronics and Telecommunications Research Institute (ETRI)
📧 [email protected]

---

📄 License

This dataset is released under the Creative Commons Attribution 4.0 International License (CC BY 4.0)

---

_{Made with ❤️ by Media Intellectualization Research Team, ETRI}