How to split the trainval data into train and validation data?

#1

Hi, I want to split the v1.0-trainval data into train and validation data. Is there any API in nuscenes-devkit can do this? And is there any tutorial about how to use eval API?

#2

Hi,

The easiest way to do it is just to load your data in such as:
X = np.array([datapoints e.g. velocity, position, rotation])
y = np.array([gt])

and then use sklearn train_test_split to split it i.e.

X_train, X_valid, y_train, y_valid = train_test_split(X,y,…)

#3

Thanks for your reply. I found default split in nuscenes/utils/splits.py, and it can be used by:
from nuscenes.utils.splits import create_splits_scenes
scene_splits = create_splits_scenes()

#4

Yes, please use create_splits_scenes() as 1) the splits should be the same as for everyone else to report comparable results and 2) you should not have samples from the same scene in different splits.