Feature output missmatch between CC & 3DMASC
Posted: Mon May 13, 2024 7:52 am
Hi,
So, I was trying again to train a classifier and left it overnight for in the morning to find a somewhat meh ressult.
So I was checking the computed feature scalar fields when I noticed the looked quite different from the onces I did compute manually in order to guess the correct scale for the classifier.
I've put together a PDF where the differences between several features can be seen, ignore the small scale differences as I've tested this not to be the root of the issue.
https://drive.google.com/drive/folders/ ... sp=sharing
In the folder there is also the .bin with the point cloud if anyone whants to check in person.
So, does 3DMASC not use CC built in feature compute and have it's own? if so, could it be made to use CC's so we can manually do tests and decide what scales fit best, since to me, CC's feature do a better job.
Cheers!
EDIT: The "Classification" SF displays wrong data, as I originally renamed OriginalCloudIndex into Classification but something whent wrong, hence the classification of 3D Masc can´t be correct due to this critical issue. Still, the mismatch on how Feature SF look remains, Yet, having the classification unproperly labelled, I can´t tell if this is a critical issue or not.
So, I was trying again to train a classifier and left it overnight for in the morning to find a somewhat meh ressult.
So I was checking the computed feature scalar fields when I noticed the looked quite different from the onces I did compute manually in order to guess the correct scale for the classifier.
I've put together a PDF where the differences between several features can be seen, ignore the small scale differences as I've tested this not to be the root of the issue.
https://drive.google.com/drive/folders/ ... sp=sharing
In the folder there is also the .bin with the point cloud if anyone whants to check in person.
So, does 3DMASC not use CC built in feature compute and have it's own? if so, could it be made to use CC's so we can manually do tests and decide what scales fit best, since to me, CC's feature do a better job.
Cheers!
EDIT: The "Classification" SF displays wrong data, as I originally renamed OriginalCloudIndex into Classification but something whent wrong, hence the classification of 3D Masc can´t be correct due to this critical issue. Still, the mismatch on how Feature SF look remains, Yet, having the classification unproperly labelled, I can´t tell if this is a critical issue or not.