Is there a way to run with a point cloud object rather than from a file?

[jalexdz]

Hi, I'm trying to run the test() function on a new single LiDAR scan in real-time without saving it as a file. From my understanding, the test() function currently loads data using get_dataset(_cfg)['test'], which seems to expect files from --dset_root.

Question: Is there a way to pass a point cloud object (with some preprocessing) directly to test() or some other method?

What I’ve Tried: I considered creating a fake dataset wrapper to mimic the expected structure—would this be a good approach? Altern

Additional Context: My goal is to run inference on live LiDAR scans in a deployed setting without saving intermediate files.

Any guidance on how to modify test() (or an alternative approach) would be appreciated! Or would it be necessary to contribute a separate predict method? Thanks for your help