Do you use rgb to generate features during preprocessing?

[LZDSJTU]

I want to train without rgb and I succeed.
I want to know that if you use rgb to generate other features when preprocessing the original data? Because if you use rgb, maybe I need to re-generate the spg and feature files. If you do not use rgb during preprocessing, I can just omit the rgb value and train.

[loicland]

We do not use the rgb values to generate the spgs for semantic3d, but use them for s3dis. You can easily change it line 156 of /partition/partition.py if need be.

What's crucial however is that you retrain the models from scratch for the inference with option --pc_attribs "xyzelpsv", as described in issue#11. Do not use our pretrained models without rgb as it will perform very badly.