About experiments.

[BenjaminYoung29]

Hi, in TABLE I you have compared RangeNet to Pointnet and others. How did you do that?

Did you simply pass the raw point cloud (.bin) into the pointnet's code?

Does that mean the RangeNet can also possess the pointnet's dataset? If RangeNet or SqueezeSeg can handle outdoor scene, I wonder their performance in indoor scene, working on the Stanford 3D semantic parsing dataset.

Thanks for your help.

[jbehley]

We reimplemented PointNet(++) and provide it a subsampled point cloud (I think 50k points).

As long as you provide a range image-like representation, then RangeNet can also process the data. We never tried the network on different then LiDAR data.