TUL-Poland/InLUT3D
This resource contains Indoor Lodz University of Technology Point Cloud Dataset (InLUT3D) - a point cloud dataset tailored for real object classification and both semantic and instance segmentation tasks. Comprising of 321 scans, some areas in the dataset are covered by multiple scans. All of them are captured using the Leica BLK360 scanner. Available categories The points are divided into 18 distinct categories outlined in the label.yaml file along with their respective codes… See the full description on the dataset page: https://huggingface.co/datasets/TUL-Poland/InLUT3D.
This resource contains Indoor Lodz University of Technology Point Cloud Dataset (InLUT3D) - a point cloud dataset tailored for real object classification and both semantic and instance segmentation tasks. Comprising of 321 scans, some areas in the dataset are covered by multiple scans. All of them are captured using the Leica BLK360 scanner.
Available categories
The points are divided into 18 distinct categories outlined in the label.yaml file along with their respective codes and colors. Among categories you will find:
- ceiling,
- floor,
- wall,
- stairs,
- column,
- chair,
- sofa,
- table,
- storage,
- door,
- window,
- plant,
- dish,
- wallmounted,
- device,
- radiator,
- lighting,
- other.
Challenges
Several challenges are intrinsic to the presented dataset:
- Extremely non-uniform categories distribution across the dataset.
- Presence of virtual images, particularly in reflective surfaces, and data exterior to windows and doors.
- Occurrence of missing data due to scanning shadows (certain areas were inaccessible to the scanner's laser beam).
- High point density throughout the dataset.
Data set structure
The structure of the dataset is the following:
inlut3d.tar.gz/
├─ setup_0/
│ ├─ projection.jpg
│ ├─ segmentation.jpg
│ ├─ setup_0.pts
├─ setup_1/
│ ├─ projection.jpg
│ ├─ segmentation.jpg
│ ├─ setup_1.pts
...