Fire360

Fire360 is a real-world dataset of around 100 high-resolution 360° videos capturing professional firefighter training exercises under diverse conditions, including day/night, indoor/outdoor, and smoke-filled environments. Each video (~2 minutes, 4K@60fps) is annotated with temporal action segments, object-level bounding boxes (gear, tools, vehicles), and environmental context tags. Unique to Fire360 is its support for studying object transformation under fire exposure, enabling tasks like damage-resilient object retrieval and episodic memory modeling. This dataset supports safety-critical vision-language research, VR training simulation, and robust multi-agent scene understanding.