Thermal anomalies (TA) dataset

This repository presents the TA dataset. This dataset is designed to support the training, evaluation, and validation of You Only Look Once (YOLO) models for thermal anomaly detection. This dataset encompasses a wide range of thermal scenarios, including controlled burns, urban heat anomalies, rural areas, and synthetic heat sources, simulating real-world conditions and potential sources of false alarms. By providing varied environmental contexts, the dataset serves as a robust foundation for developing and refining thermal anomaly detection models, especially under challenging and dynamic conditions.

Dataset Composition:
The TA dataset comprises approximately 4,432 thermal images sourced from multiple previous measurement campaigns and state-of-the-art open datasets, such as Advanced Driver Assistance Systems (ADAS) and TarDAL M3FD. The dataset is categorized as follows:

External Datasets: Includes thermal images from ADAS and M3FD, focused on urban environments.
In-House Data: Thermal images captured using various FLIR and Seek cameras in urban, rural, and wilderness settings.

Technical Specifications:
File Format: JPEG, TXT, NPY
Resolution: Ranges from 160x120 to 640x512
Cameras Used: FLIR Tau 2, FLIR A615, FLIR A35, Seek Mosaic, and FLIR Lepton

Attached to this repository is the dataset created and used during the work. If you wish to use the code sample to replicate our work, please use the link found in the 'Related links' section, which leads directly to the official GitHub of the paper.