How to Run ScanNet

January 27, 2023 · View on GitHub

Download Dataset

  1. Request to download the dataset as mentioned in https://github.com/ScanNet/ScanNet.
  • If the request is accepted, they will send you a script called download-scannet.py.
  1. Place the file download-scannet.py under scripts/.
  2. Run preprocess-scannet.sh and you should be ready to run ScanNet on Orbeez SLAM. Note: Please run the script under scripts due to the dicrectory structure.

The directory structure would be as follow

└─📁 scans/
  ├──📁 scene0000_00/
  │ ...
  ├─📁 scene0059_00/
  │ ├─📁 depth/
  │ ├─📄 trajectory.txt
  │ ├─📄 scene0059_00.sens
  │ ├─📁 pose/
  │ └─📁 color/
  ├─📁 scene0207_00/
  │ ...
  ├─📁 scene0169_00/
  │ ...
  ├─📁 scene0181_00/
  │ ...
  ├─📁 scene0106_00/
  │ ...

Run Guideline

RGB-D

The main program is in Examples/RGB-D/rgbd_scannet.cu Usage:

rgbd_scannet path_to_vocabulary path_to_settings path_to_sequence
  • Run scene0000_00
./build/rgbd_scannet ./Vocabulary/ORBvoc.txt ./configs/RGB-D/ScanNet/scene0000_00.yaml /data/scene0000_00/
  • Run scene0059_00
./build/rgbd_scannet ./Vocabulary/ORBvoc.txt ./configs/RGB-D/ScanNet/scene0059_00.yaml /data/scene0059_00/
  • Run scene0106_00
./build/rgbd_scannet ./Vocabulary/ORBvoc.txt ./configs/RGB-D/ScanNet/scene0106_00.yaml /data/scene0106_00/
  • Run scene0169_00
./build/rgbd_scannet ./Vocabulary/ORBvoc.txt ./configs/RGB-D/ScanNet/scene0169_00.yaml /data/scene0169_00/
  • Run scene0181_00
./build/rgbd_scannet ./Vocabulary/ORBvoc.txt ./configs/RGB-D/ScanNet/scene0181_00.yaml /data/scene0181_00/
  • Run scene0207_00
./build/rgbd_scannet ./Vocabulary/ORBvoc.txt ./configs/RGB-D/ScanNet/scene0207_00.yaml /data/scene0207_00/

Monocular

The main program is in Examples/Monocular/mono_scannet.cu Usage:

mono_scannet path_to_vocabulary path_to_settings path_to_sequence

Note that monocular does not perform well and we did not find a config ('path_to_settings') to make it work. Therefore, we do not provide config files for mono_scannet under configs/Monocular. If you want to run the monocular setting, you can try to put the RGB-D config path in the 'path_to_settings' to see the outcome.

  • i.e.: Change the build/rgbd_scannet above to build/mono_scannet.

Acknowledgement

SensorData.py and reader.py are modified from https://github.com/ScanNet/ScanNet, where we convert them from Python 2 to Python 3.