University-1652 is released for NON-COMMERCIAL ACADEMIC RESEARCH ONLY. By requesting access, you agree to the full Terms of Use in the dataset card, including: (1) no commercial use of the dataset; (2) no redistribution of the dataset or any part of it — models, code and evaluation results derived from it may be published; (3) citing the paper "University-1652: A Multi-view Multi-source Benchmark for Drone-based Geo-localization" (ACM MM 2020) in any resulting publication; (4) the imagery originates from Google Maps/Earth and all associated rights remain with Google LLC and its content providers — you agree to use it only for the benchmark tasks defined in the publication and in compliance with the original providers' terms; (5) no use for identifying individuals, surveillance of individuals, or any privacy-violating application; (6) deleting affected content upon a takedown notice, and deleting all copies if access is revoked. Access is granted to the individual applicant — each member of a lab or team should submit their own request.

University-1652: Drone-based Geo-localization Benchmark 🚁

University-1652 is a multi-view dataset for drone-based geo-localization, annotating 1652 buildings across 72 universities (ACM Multimedia 2020, paper). Cited in 500+ papers, it supports Drone → Satellite localization and Satellite → Drone navigation.

🔗 Official code & baseline: layumi/University1652-Baseline · Leaderboard: State-of-the-art results

🔐 Access

This dataset is gated: click "Request access" above, fill in the short form and agree to the research-only terms, and access is granted automatically. (This replaces the previous email-based request process — the Hugging Face route is now the recommended and fastest way.)

📊 Dataset Structure

Splits:
- Train: 50,218 images (drone, satellite, street, google; 33 universities)
- Test:
  - query_drone: 37,855 images
  - gallery_drone: 51,355 images
  - query_street: 2,579 images
  - gallery_street: 2,921 images
  - query_satellite: 701 images
  - gallery_satellite: 951 images
  - 4K_drone: 12 images
Features:
- image: Drone/satellite/street/4K_drone images
- building_id: Building identifier
- view_type: drone/satellite/street/drone_4k
- split_type: train/query/gallery
Size: ~9.2GB (unzipped)

🚀 Usage (Download Only)

Since University-1652 is a standard multi-view vision benchmark with a strict directory structure, this repository is intended for raw file downloading only. You do not need to load it via datasets.load_dataset(). Instead, download the raw folders and plug them directly into your standard PyTorch DataLoader (e.g., torchvision.datasets.ImageFolder).

Note: Because access is gated, you must be logged in to Hugging Face before downloading. Run huggingface-cli login (or set the HF_TOKEN environment variable) with a token from an account that has been granted access.

Method 1: Using Hugging Face Python API (Recommended)

This is the most reliable way to download the entire dataset with proper resume-on-failure support.

# pip install huggingface_hub
# huggingface-cli login   <- required once, after your access request is approved
from huggingface_hub import snapshot_download

# Download the dataset and keep the original folder structure (train/ and test/)
local_dir = snapshot_download(
    repo_id="layumi/university-1652",
    repo_type="dataset",
    local_dir="./University-1652",  # Your local destination directory
    local_dir_use_symlinks=False    # Download actual files instead of symlinks
)
print(f"Dataset successfully downloaded to: {local_dir}")

Method 2: Git Clone

Ensure you have Git LFS installed on your system before cloning, and that your git credentials are configured with your Hugging Face token.

git lfs install
git clone https://huggingface.co/datasets/layumi/university-1652

⚠️ Terms of Use & Legal Disclaimer

1. Research Purpose Only

This dataset is provided strictly for non-commercial academic research purposes. Any commercial utilization, including but not limited to training commercial models, commercial image retrieval systems, or integration into proprietary software, is strictly prohibited without prior written permission from the authors.

2. No Redistribution

You may not distribute, publish, or share the dataset (in whole or in part), or any derivative datasets, with any third party. Others who wish to use the dataset must request access through this page. You may publish models, code, and evaluation results derived from the dataset.

3. Attribution

Any publication or public work that uses this dataset must cite the University-1652 paper (see Citation below).

4. Google Maps Data Fair Usage & Claim

Data Origin & Ownership: This dataset contains imagery crawled from and property of Google Maps/Earth. All copyright, intellectual property rights, and trademarks associated with these images remain the sole property of Google LLC and its content providers.
Fair Use Notice: The incorporation of these images into University-1652 is intended solely for academic benchmarking and scientific advancement under "Fair Use" guidelines.
No Endorsement: This dataset is an independent academic project and is not endorsed, sponsored, or affiliated with Google LLC.
Redistribution: Users are prohibited from isolating, extracting, or utilizing the Google Maps imagery for any purposes outside the benchmark tasks defined in the official publication.

5. Privacy and Ethics

You will not use the dataset to identify or infer information about specific individuals, and will not use it for surveillance of individuals or any application that violates privacy or human rights.

6. Takedown Policy

If any content in the dataset is subject to a valid removal request, you agree to delete the affected content upon notification by the authors. To report an issue, please open an issue on the GitHub repository or contact the authors.

7. No Warranty & Termination

The dataset is provided "as is" without warranty of any kind. Access may be revoked at any time if these terms are violated; upon revocation, you agree to delete all copies of the dataset.

🏆 Citation

If you find this dataset or the benchmark helpful for your research, please consider citing the original paper:

@inproceedings{zheng2020university,
  title={University-1652: A Multi-view Multi-source Benchmark for Drone-based Geo-localization},
  author={Zheng, Zhedong and Wei, Yunchao and Yang, Yi},
  booktitle={Proceedings of the ACM International Conference on Multimedia (ACM MM)},
  year={2020}
}

🔗 Related Resources

💻 Baseline code & pretrained models: University1652-Baseline
📈 Leaderboard / state-of-the-art: Awesome-Geo-localization
📝 Text-guided extension: GeoText-1652

Downloads last month: 15,666

Total file size:

9.21 GB

Paper for layumi/university-1652

University-1652: A Multi-view Multi-source Benchmark for Drone-based Geo-localization

Paper • 2002.12186 • Published Feb 27, 2020