ONFIRE Dataset

The Fire Detection Video Dataset is a collection of 322 videos specifically curated to address the challenges associated with fire detection in diverse conditions. This dataset is unparalleled in its heterogeneity, encompassing variations in image resolution, illumination, distance from fire or smoke, pixel size of flame or smoke, background activity, and the scenario (urban or wildfire). Moreover, it stands out as the most extensive and diverse fire detection video dataset, with annotations that include the fire ignition time – a unique feature not commonly found in public datasets.