Polyvore Outfits Dataset

📄 Paper 💻 Code 📬 Contact

📌 Access Requirements

Access to this dataset is restricted and intended solely for academic and non-profit research purposes. By requesting and downloading this dataset, you acknowledge and agree to the following:

1. Approval required: Access should be requested using an institutional or research-affiliated email address.
2. Copyright notice: The dataset authors do not own the copyright to the underlying images or metadata. All content is provided for research and educational use.
3. Requesting access: Researchers wishing to use the dataset should request access through the "Access Request" button on this official page.

📖 Dataset Summary

The Polyvore Outfits Dataset was introduced in:

Learning Type-Aware Embeddings for Fashion Compatibility Mariya I. Vasileva, Bryan Plummer, Khrisna Dusad, Shreya Rajpal, Ranjitha Kumar, David A. Forsyth European Conference on Computer Vision (ECCV), 2018

The dataset was designed to support research in fashion compatibility, outfit recommendation, and multimodal retrieval. It enables evaluation of models that reason about complementarity between clothing items across semantic categories.

📝 Key Characteristics

• 68,306 curated outfits
• 261,058 heterogeneous fashion items
• High-quality product images
• Rich metadata, including:
  • Category labels
  • Item descriptions
  • Outfit groupings
• Widely adopted as a benchmark for compatibility and retrieval tasks

🧩 Dataset Versions & Evaluation Splits

Two official variants are provided:

1. Polyvore Outfits (Nondisjoint)

• Easier, outfit-level split
• No outfit overlap between training, validation, and test sets
• Individual items may appear in both training and test sets

2. Polyvore Outfits (Disjoint)

• Harder, item-level split
• No item overlap between training, validation, and test sets
• Recommended for fair generalization assessment

When reporting experiment results, users should specify which split was used for evaluation.

🎯 Supported Tasks

• Fashion compatibility prediction
• Outfit recommendation
• Fill-in-the-blank (FITB) evaluation
• Outfit and item retrieval
• Visual relationship mining
• Contrastive and metric learning

---

⚖️ License & Terms of Use

This dataset is distributed under the Creative Commons Attribution 4.0 International (CC BY 4.0) license. By downloading or using this dataset, you agree to the following terms:

1. Commercial & non-commercial use permitted You are free to use, copy, and redistribute the material in any medium or format for any purpose, including commercial applications.

2. Attribution required You must give appropriate credit by citing the original paper (see "How to Cite" below), provide a link to the license, and indicate if and how changes were made.

3. Adaptation permitted You may remix, transform, and build upon the material, provided the original work is properly attributed.

4. No additional restrictions You may not apply legal terms or technological measures that legally restrict others from doing anything the license permits.

---

📚 How to Cite

If you use this dataset, build upon it, or include it in a commercial application, please cite the original work:

@inproceedings{vasileva2018learning,
  title={Learning Type-Aware Embeddings for Fashion Compatibility},
  author={Vasileva, Mariya I. and Plummer, Bryan A. and Dusad, Krishna and Rajpal, Shreya and Kumar, Ranjitha and Forsyth, David A.},
  booktitle={European Conference on Computer Vision (ECCV)},
  year={2018}
}

---

📬 Contact

For questions regarding access, attribution, or permitted use, please contact the dataset authors here.