SemanticKITTI Validation Set Results

[alextanned]

Hello, thank you for publishing this repository, your work is very insightful.

Regarding Table 1. in the paper about validation set comparisons on SemanticKITTI, may I ask how you obtained the mIOU results of the other methods? I know that you evaluate on your own dataset created by create_fov_dataset.py , but do you:

1. Use the published pretrained checkpoint from those methods (e.g. Salsanext, Cylinder3D, etc.) to evaluate directly even though most of them are trained on the entire point cloud (360 degree fov). Or,
2. Use the published model architecture and training settings of those methods and train yourselves using the limited fov dataset

If none of the above is correct, it would be much appreciated if you give me some insights on how you conducted the comparison. Thanks again for your great work!