Cityscapes-VPS Video Panoramic Segmentation Dataset

Cityscapes-VPS is a video extension of the Cityscapes dataset validation set. This dataset provides 2,500 panoptic labels, which expands the original 500 image panoptic labels of the Cityscapes dataset to 3,000. These panoptic labels correspond to the 5th, 10th, 15th, 20th, 25th, and 30th frames of every 500 videos, where all instance IDs change over time.

This dataset not only supports the video panoptic segmentation (VPS) task, but also provides super-set annotations for the video semantic segmentation (VSS) and video instance segmentation (VIS) tasks.