what is divide task ["Vehicle", "Pedestrain", "Cyclist"] -> ["Vehicle"] , ["Pedestrain", "Cyclist"]

[lejk8104]

thanks your fully support about PIllarNet

I am conduct into training PillarNet model reference of your github page. (same config and setting)
but some difference about CenterPoint-voxel config file.

pillarnet path: /configs/waymo/voxelnet/waymo_centerpoint_pillarnet_3x.py
tasks = [ dict(stride=8, class_names=['VEHICLE']), dict(stride=8, class_names=['PEDESTRIAN', 'CYCLIST'])] class_names = list(itertools.chain(*[t["class_names"] for t in tasks]))

centerpoint path: configs/waymo/voxelnet/waymo_centerpoint_voxelnet_3x.py
tasks = [ dict(num_class=3, class_names=['VEHICLE', 'PEDESTRIAN', 'CYCLIST']),] class_names = list(itertools.chain(*[t["class_names"] for t in tasks]))

1. why is difference about config file?
2. GPU mermory changed frequently during training Pillarnet
   
   
   
   it is general case? is not negative effect about Pillarnet accuracy? or training?

[WWDCRobot]

You can also use the same config without division for the same results
The division is for the next work