In your paper, A Benchmark for LiDAR-based Panoptic Segmentation based on KITTI, it says, " We provide temporally-consistent instance annotations for all traffic participants including vehicles, pedestrians, bicyclists, and motorcyclists for the KITTI Odometry Benchmark." How do we understand "temporally-consistent"? For example, does one instance have identical 32 bit labels in sequential frames?
Yes, That's the case. We tried to ensure that instances of the same class are the same over the whole sequence. No matter if they are non-moving or moving.
In your paper, A Benchmark for LiDAR-based Panoptic Segmentation based on KITTI, it says, " We provide temporally-consistent instance annotations for all traffic participants including vehicles, pedestrians, bicyclists, and motorcyclists for the KITTI Odometry Benchmark." How do we understand "temporally-consistent"? For example, does one instance have identical 32 bit labels in sequential frames?