Automate point cloud classification for multiple tiles

[vincen]

Hello again,

I was wondering if there was a way to automate the classification of 158 SfM point cloud tiles, either with 3DMASC or CANUPO.

There is about 1 billion points in my original point cloud so I do not have a choice but to do some tiling.

Is one training efficient for the other tiles ?

Thank You

Vincent

[daniel]

I'm not the expert, but I believe as long as the other tiles have the same density and objects of the same scale, then the classifier should work on these tiles as well.

And you can normally automate the process via the command line.

[paul.leroy]

This is one thing that really matters in the world of classification: "does my classifier generalizes well?".

If so, you are a happy machine learning user.

If not, you will have to investigate and find the reason why samples not seen during the training phase are not well classified.
You may have to rework your training sample set to cover the variety of configurations. I would recommend to select samples not only in one tile but rather pick up relevant samples in an area larger than a single tile otherwise you will miss the diversity of your overall dataset. But it really depends on your specific problem. If tiles are very similar, a classifier trained on one tile should generalize on the others.