Table 1 ScanNet++ split in Sonata Paper

Great work on exploring self-supervised learning with pure point cloud data! I was reading your paper and noticed that the ScanNet++ statistics in Table 1 don’t seem to match the version of the dataset I have. From what I know, version 2 should include 856 training scenes, 50 validation scenes, and 50 test scenes. Could you clarify if I’m mistaken or if you used a different split?


Start by reading Table 1 in the Sonata paper and checking which ScanNet++ dataset version and split it reports. Compare those figures with the stated version 2 counts of 856 training, 50 validation, and 50 test scenes; done means clarifying the discrepancy or correcting the reported statistics. No files or tests are named in the issue.

Written by the indexing model from the issue text.