Leader360V Dataset

Leader360V is the first large-scale (10K+), labeled real-world 360 video datasets for instance segmentation and tracking. Our datasets enjoy high scene diversity, ranging from indoor and urban settings to natural and dynamic outdoor scenes.

All videos in this dataset have undergone standardized preprocessing, including video clipping, facial anonymization for privacy protection, and balanced scene distribution across categories.

Regarding dataset composition, we integrated existing 360 video datasets (either unlabeled or annotated for single tasks) and supplemented them with newly collected self-recorded videos. All content was then re-annotated to support joint segmentation and tracking tasks.

Following paper acceptance, we will publicly release all code and datasets for free use by researchers in academia and industry.