We propose the first large-scale and real-world 360 dataset and a automatic annotate pipeline to reduce the cost of manual annotation.