Dealing with labels in Accumulated Point Clouds

[vhchuong]

Hello! I'm working with the Accumulated Point Clouds version of kitti-360. This version contains annotated 45 classes. I am baffled about how to process these classes to create trainable instances and/or ignore some. Can anyone give some insights or instructions for this type of dataset? Thank you so much !

[yiyiliao]

Hi, for creating trainable semantic labels, you can use the id2trainId function to convert the original semantic labels to trainId.

For creating trainable instance labels to submit to our evaluation server, you can keep building and car and ignore the other classes, as we only consider these two classes in our evaluation script.