How to structure the input data?

[Winterruhe]

Hi there,

I'm relatively new to deep learning and a bit confused on how to structure my point cloud for semantic segmentation. I have tried creating batches of my point cloud using next neighbour search and using all points in one batch. In both cases the segmentation looks quite random.

Maybe I'm missing the basics to understand this code, but is there any further information or documentation that could help me?

Thanks in advance.