No reachable cells

[drStacky]

I'm trying to run my own point clouds through the algorithm. The simplest is a box shape (with floor).

The conversion into `numpy` binary files seems to work. I computed planes using Mapple as instructed, and the code appears to find them.

number of planes: 7
number of cells: 22

However, I get no output .obj file.

cut performed: 0.00 s
cut_value: 0.00
number of extracted cells: 0

For reference, PolyFit works on the same .vg file.

I would appreciate any help troubleshooting this issue. Thanks!

[chenzhaiyu]

Hi, the checkpoint was trained on data that is quite different from yours. You will need to retrain a model on your own data. See https://github.com/chenzhaiyu/points2poly/issues/18 and https://github.com/chenzhaiyu/points2poly/issues/28.

[drStacky]

Thank you for the fascinating work and quick reply. What does that mean that the data is quite different? I dug into the .xyz.npy files (that I believe are the inputs into the SDG NN), and the values are in very similar ranges to the samples given, so it doesn't seem like a normalization issue. I'd love to try this method out on some real-world data without having to make my own labeled training set.

[drStacky]

FWIW The values coming out of the NN also seem reasonable, at least in terms of their range. For example: min, max = -18.1156, 7.8570.

[chenzhaiyu]

Hi, by "data quite different from yours" I mean simply the geometry (point distribution), not the file format. Your sample has an overreaching bottom (i.e., an overhanging base) that the training set did not include, which could reasonably lead the network to misclassify interior and exterior cells, even if you processed the data correctly.