🧵 EquiFashion-DB (Mini)

EquiFashion-DB (Mini) is a compact subset of the full EquiFashion-DB dataset, designed for text-to-image and pose-guided fashion generation.
It supports fast experimentation for GAN–Diffusion hybrid models and Stable Diffusion fine-tuning.

📂 Structure

EquiFashion-DB/
├── train/               # 4,000 fashion model images + captions
├── test/                # 1,208 fashion model images + captions
├── train_pose/          # 4,680 human pose reference maps
├── train.json           # captions for training images
├── test.json            # captions for test images
└── train_pose.json      # keypoints for pose annotations

🧠 Annotation Format

Text captions

{
  "image": "train/000123.jpg",
  "caption": "Blue floral long-sleeve dress with tied waist."
}

Pose annotations

{
  "image": "train_pose/pose_00145.png",
  "keypoints": [[x1, y1], [x2, y2], ..., [x17, y17]]
}

⚙️ Summary

Field	Description
Total images	~9,888
Resolution	512×512 px
Languages	English (and bilingual captions available)
License	MIT
Tasks	Text-to-Image, Pose-Guided Generation

🧩 Usage

from datasets import load_dataset
dataset = load_dataset("NguyenDinhHieu/EquiFashion-DB")

📖 Citation

@dataset{NguyenDinhHieu2025EquiFashionDB,
  title   = {EquiFashion-DB (Mini): Multimodal Dataset for Fashion Generation},
  author={},
  booktitle={Proceedings of the ... Conference},
  year={2025},
  organization={FPT University, Hanoi}
}