search

Jun-CEN / Open-world-3D-semantic-segmentation

[ECCV 2022] Open-world Semantic Segmentation for LIDAR Point Clouds
67 stars 10 forks source link

Change in semantic-kitti-upper.yaml #4

Closed ChungInchul closed 1 year ago

ChungInchul commented 1 year ago

Hi, Jun Cen. First of all thanks for the great work.

I have a small question.

In line 135 of semantic-kitti.yaml 'other-object' was mapped to 'unlabeled' as it is usually ignored for training or evaluation.

However, in line 135 of semantic-kitti-upper.yaml used for training upper bound, 'other-object' was mapped to 'other-vehicle'.

Is there any reason for that?

Jun-CEN commented 1 year ago

Hi,

It seems like we should set 'other-object' as 'unlabeled' for the upper bound. I am a little confused about why I changed the setting like that. But it should not influence the final result with a large margin. Upper bound simply means we train the model for the all classes of SemanticKITTI.

ChungInchul commented 1 year ago

Thanks!