about validation on nuscenes

[Terminal-K]

about validation on nuscenes

[Terminal-K]

Many thanks!! You really help me a lot!

[benemer]

The sequence parameter refers to the sequence ID in nuscenes. For example, for sequence scene-0271, you have to pass the sequence ID 271. You can find the IDs for the validation set here or get it via Python:

from nuscenes.utils.splits import val
sequences = [v.split("-")[1] for v in val]

Note that you have to modify these lines in the Nuscenes data loader to load the correct database. Be aware that calling the NuScenes function in the data loader takes some time for the "v1.0-trainval" data and you have to do it for each sequence of the validation set. To mitigate this, you can modify the pipeline to iterate over all sequences and modify the data loader to create the NuScenes object once on import:

[..]
import nuscenes
from nuscenes.nuscenes import NuScenes

__nusc_dataroot__ = "/path/to/nuScenes"
__nusc_version__: str = "v1.0-trainval"
 __nusc__ = NuScenes(dataroot=__nusc_dataroot__, version=__nusc_version__)

class MOSNuscenesDataset(NuScenesDataset):
[...]

And then you set self.nusc = __nusc__ in the __init__.