Caltech Aerial RGB-Thermal Dataset in the Wild

Description

Summary

This is the first publicly available RGB-thermal dataset tailored for aerial field robotics operating in natural environments. It encompasses various terrains across the continental United States, including rivers, lakes, coastlines, deserts, and forests. The dataset captures synchronized RGB, long-wave thermal, GPS, and IMU data.  Additionally, semantic segmentation annotations for 10 classes commonly encountered in natural settings are provided to aid in the development of thermal perception algorithms. 

Benchmarks

The dataset introduces benchmarks for:  thermal semantic segmentation, RGB-Thermal semantic segmentation, RGB-to-Thermal image translation, and visual-inertial odometry (VIO) / simultaneous Localization and Mapping (SLAM). 

Code

Please use the accompanying code and scripts to help process the data: https://github.com/aerorobotics/caltech-aerial-rgbt-dataset 

Citation: 

If this dataset has been useful to you, please cite our paper:

@article{lee2024cart,  title={CART: Caltech aerial RGB-thermal dataset in the wild},  author={Lee, Connor and Anderson, Matthew and Raganathan, Nikhil and Zuo, Xingxing and Do, Kevin and Gkioxari, Georgia and Chung, Soon-Jo},  journal={arXiv preprint arXiv:2403.08997},  year={2024} }

Files

External Files

Files available via S3 at https://renc.osn.xsede.org/ini210004tommorrell/cks6g-ps927/data/

Additional details