1) We prepared a preprocessed version of ScanNetv2 dataset. Download it at GoogleDrive
2) Extract the downloaded zip file to geoformer/data as follows.
geoformer
├── data
│ ├── scannetv2
│ │ ├── scenes
│ │ ├── val_gt
│ │ ├── support_sets
│ │ ├── class2instances.pkl
│ │ ├── class2scans.pkl
│ │ ├── test_combinations_fold0
│ │ ├── test_combinations_fold1
│ │ ├── scannetv2_train.txt
│ │ ├── scannetv2_test.txt
│ │ ├── scannetv2_val.txt
...
where scenes is the preprocessed 3d point cloud data stored in .npy format, val_gt stores groundtruth instance masks, support_sets stores the predefined support sets (used to eval all experiments)
3) If you cannot download our preprocessed data, you can download the original ScanNetv2 dataset from ScanNet. Put all the files in geoformer/data/scannetv2/raws folder and run two scripts data/scannetv2/prepare_data_inst.py and data/scannetv2/prepare_data_inst_gt.py.