Thermal-UAV Dataset

Thermal-UAV is a comprehensive dataset for cross-modal thermal geo-localization, introduced in the paper SCC-Loc: A Unified Semantic Cascade Consensus Framework for UAV Thermal Geo-Localization.

It is designed for Unmanned Aerial Vehicles (UAVs) operating in GNSS-denied environments, addressing the modality gap between thermal queries and visible-light satellite imagery.

• Paper: SCC-Loc: A Unified Semantic Cascade Consensus Framework for UAV Thermal Geo-Localization
• GitHub Repository: FloralHercules/SCC-Loc

Dataset Description

The Thermal-UAV dataset comprises 11,890 thermal UAV images captured using a DJI Matrice 4T drone around Changsha. These images feature nadir views, multi-temporal and multi-scenario data, and continuous flight sequences.

The dataset is split into:

• Training: 8,115 images
• Validation: 1,425 images
• Testing: 2,350 images

Additionally, the dataset provides corresponding Google Maps satellite imagery at 0.26 m/px and a Digital Surface Model (DSM) at 5.29 m/px.

Dataset Structure

The directory structure of the Thermal-UAV dataset is as follows:

Data/
├── metadata/
│   ├── train_Thermal.json
│   ├── valid_Thermal.json
│   └── test_Thermal.json
├── Reference_map/
│   └── changsha/
│       ├── ref.tif
│       └── dsm.tif
└── Thermal-UAV/
  ├── train/
  │   └── changsha/
  │       └── <NUDT>/
  │           └── Thermal/
  │               ├── xxx1.JPG
  │               ├── xxx2.JPG
  |	          └──  Thermal_info.csv
  │       └── ...
  ├── valid/
  └── test/

Metadata Format

Each thermal query in the metadata JSON files follows this structure:

{
  "name": "./Data/Thermal-UAV/train/changsha/city_300_ortho_night/Thermal/xxx_T.JPG",
  "lat": 28.2436611,
  "lon": 112.9985009,
  "abs_height": 341.354,
  "rel_height": 323.967,
  "pitch": -90.0,
  "yaw": 88.0,
  "roll": 180.0,
  "cam_size": 9.83,
  "focal_len": 12.0,
  "width": 1280.0,
  "height": 1024.0
}

Citation

If you find this dataset or the SCC-Loc framework useful for your research, please consider citing:

@article{sccloc2026,
  title={SCC-Loc: A Unified Semantic Cascade Consensus Framework for UAV Thermal Geo-Localization},
  author={Xiaoran Zhang, Yu Liu, Jinyu Liang, Kangqiushi Li, Zhiwei Huang, Huaxin Xiao},
  journal={arXiv:2604.03120},
  year={2026}
}

License

This dataset is released under the Apache License 2.0. Please note that the data is intended for academic research purposes.