Datasets:

s7er
/

ChangAnPark

pretty_name: ChangAnPark
task_categories:
  - other
tags:
  - image-retrieval
  - visual-localization
  - uav
  - cross-domain
  - 6dof-pose-estimation
size_categories:
  - 10K<n<100K
license: cc-by-nc-4.0
gated: true
extra_gated_heading: Request access to ChangAnPark
extra_gated_description: Requests are manually reviewed for non-commercial use.
extra_gated_button_content: Submit access request
extra_gated_prompt: >-
  Please provide accurate affiliation and intended-use information. Access is
  governed by the Creative Commons Attribution-NonCommercial 4.0 International
  license (CC BY-NC 4.0).
extra_gated_fields:
  Full name: text
  Affiliation: text
  Institutional email: text
  Country: country
  Intended non-commercial use: text
  Requested components:
    type: select
    options:
      - Retrieval package only
      - Retrieval and PnP coordinate-map packages
  I will comply with the dataset license: checkbox

ChangAnPark

ChangAnPark is the self-collected Chang'an Park benchmark used to evaluate retrieval, geometric re-ranking, and absolute PnP localization in RIM: A Retrieval-In-Matching Framework for Cross-Domain Global Visual Localization of UAVs.

Files

File	Contents	Required for
`ChangAnPark-v0.1.0-retrieval.tar.zst`	26,568 SfM reference RGB images, 2,383 real UAV queries, public poses, configuration, and the 68-query exclusion protocol	Retrieval and re-ranking
`ChangAnPark-v0.1.0-coordinate-maps.tar.zst`	26,568 per-pixel float32 ECEF coordinate maps	PnP
`SHA256SUMS.txt`	SHA-256 checksums for both archives	Integrity verification
`LICENSE_DATASET.md`	Dataset license and ownership notice	Terms of use

Retrieval-only evaluation needs only the retrieval archive. Extract both archives into the same parent directory for the complete layout:

sha256sum -c SHA256SUMS.txt
tar --zstd -xf ChangAnPark-v0.1.0-retrieval.tar.zst
tar --zstd -xf ChangAnPark-v0.1.0-coordinate-maps.tar.zst

ChangAnPark-v0.1.0/
├── DATASET_CARD.md
├── LICENSE_DATASET.md
└── park/
    ├── configs/park.json
    ├── protocols/park_construction_excluded_indices.json
    ├── train/
    │   ├── poses_public.npy
    │   ├── rgb/
    │   └── coords_3d/
    └── test/
        ├── poses_public.npy
        └── rgb/real/

The paper protocol evaluates 2,315 queries after excluding the 68 indices in the included protocol JSON. Full-ECEF coordinate maps are converted to the configured local frame by the RIM evaluator immediately before PnP.

Access

Access requests are reviewed manually. Include your name, affiliation, institutional email, country, intended non-commercial use, and requested components in the form above. Questions may be sent to lixin200@mails.ucas.edu.cn.

Gating controls initial access to the files. Once access is granted, use and sharing are governed by CC BY-NC 4.0, including its attribution, non-commercial-use, and indication-of-changes conditions.

License and ownership

Except where otherwise noted, ChangAnPark is licensed under the Creative Commons Attribution-NonCommercial 4.0 International license (CC BY-NC 4.0). Copyright and database rights remain with the dataset authors; the license does not transfer ownership. Commercial use is not licensed. For commercial licensing enquiries, contact lixin200@mails.ucas.edu.cn.

This repository contains no Google-derived imagery, Google 3D Tiles, model weights, experimental feature caches, or raw Park 3D Tiles/B3DM assets. The RIM source code and model weights have their own licenses.

Citation

Please cite the paper associated with this dataset:

Xin Li, Siyuan Duan, Shang Wang, Zhimin Mao, Bingliang Hu, and Geng Zhang. RIM: A Retrieval-In-Matching Framework for Cross-Domain Global Visual Localization of UAVs. Manuscript submitted to Knowledge-Based Systems, 2026.

Complete publication metadata and BibTeX will be added when finalized.