FLAME stands for Fire Luminosity Airborne-based Machine learning Evaluation. It is a forest fire detection dataset based on aerial images, which is used to promote forest fire monitoring and alarm. This dataset is a fire image dataset collected by drones during the prescribed burning of piled debris in the pine forests of Arizona. The dataset includes video records and heat maps taken by infrared cameras. The captured videos and images are annotated frame by frame, which can help researchers easily build models.