Training phase : classification of the point cloud

[Madeleine]

As I understand, for the training phase of 3DMASC, we need a classified point cloud.

In his tutorial videos, M. Kharroubi (previous post: https://www.linkedin.com/pulse/automati ... bderrazzaq) uses a cloud already well classified, to train his classifier. This way, the classifier can know if the points predicted match the real points. In real life, we need a classifier especially because we need to classify a "raw data" point cloud.
Therefore, if we don't have a classified point cloud of our region of interest, can we proceed in the same way we did with the CANUPO plugin and select just a few bits of the point cloud, assigning those to vegetation or ground (let's say my goal is to remove vegetation)?

Thank you for the feedback,

Madeleine

[paul.leroy]

Exactly, you need a labelized point cloud to train your classifier. As a rule of thumb, 2000 points per class are sufficient. Be careful to select points which represent the diversity of each class: if is is vegetation, pick points in various types of vegetation, in different areas. Also, take care of the balance between classes.