Question aboud dynamic class

[jimchenhub]

Thanks for you work. I have a question about the setting of dynamic class. In your code, dynamic classes are

dynamic_class = [0, 1, 3, 4, 5, 7, 9, 10]

According to the label mapping as below, that means noise and barrier are dynamic classes and bicycle and motorcycle are not dynamic classes. This setting seems a bit unreasonable. What is the reason for this setting?

label_map = {
    'animal':0, 
    'human.pedestrian.personal_mobility':0, 
    'human.pedestrian.stroller':0, 
    'human.pedestrian.wheelchair':0, 
    'movable_object.debris':0,
    'movable_object.pushable_pullable':0, 
    'static_object.bicycle_rack':0, 
    'vehicle.emergency.ambulance':0, 
    'vehicle.emergency.police':0, 
    'noise':0, 
    'static.other':0, 
    'vehicle.ego':0,
    'movable_object.barrier':1, 
    'vehicle.bicycle':2,
    'vehicle.bus.bendy':3,
    'vehicle.bus.rigid':3,
    'vehicle.car':4,
    'vehicle.construction':5,
    'vehicle.motorcycle':6,
    'human.pedestrian.adult':7,
    'human.pedestrian.child':7,
    'human.pedestrian.construction_worker':7,
    'human.pedestrian.police_officer':7,
    'movable_object.trafficcone':8,
    'vehicle.trailer':9,
    'vehicle.truck':10,
    'flat.driveable_surface':11,
    'flat.other':12,
    'flat.sidewalk':13,
    'flat.terrain':14,
    'static.manmade':15,
    'static.vegetation':16
}

[pmj110119]

Thanks for pointing it out.

It's been a while, but I remember that the reason for this setting is that in the nuscenes dataset, most of bicycles and motorcycles are "parked" on the roadside, so I classified them as static objects.

[jimchenhub]

Thanks for your quick reply. Parked bicycles seem reasonable as static class. But the barrier are generally not move. To make an accurate labeling, it might be a good choice to use the original categories of LiDAR point.

[pmj110119]

Yes, what you said is a more reasonable way.