scannet

ScanNet dataset

ScanNet is an RGB-D video dataset containing 2.5 million views in more than 1500 scans, annotated with 3D camera poses, surface reconstructions, and instance-level semantic segmentations. For more details, see: ScanNet v2

Note: 3D meshes and surface reconstructions are not included in the preparation of the dataset.

Prepare dataset

1. Download the Dataset:

   To be able to download the dataset fill out the ScanNet Terms of Use and send it to them at [email protected]. Once your request is approved, you will receive a download_scannet.py script.

   Execute it with:

   # general usage
   python download-scannet.py -o /path/where/to/download/ScanNet

2. Convert dataset:

   # general usage (note that one process might use more than 3GB RAM)
   nicr_sa_prepare_dataset scannet \
       /path/where/to/download/ScanNet \
       /path/where/to/convert/ScanNet \
       [--n-processes N] \
       [--subsample N0]
       [--additional-subsamples N1 N2]
       [--label-map-file /path/to/scannet-labels.combined.tsv]

With arguments:

• --n-processes: The number of worker processes to spawn.
• --subsample The subsample that is exported to the output folder.
• --additional_subsamples: Tor additional subsampled versions of the dataset.
• --label-map-file: Path to scannet-labels.combined.tsv, if not specified assumed to be located in source dir.