Tools

We provide annotation data for both RGBD and LIDAR data. Each point from image or point cloud is segmented as belonging to one of the planes or being not planar.

Annotation for RGBD data is represented by colored images with the same size as the raw depth images from the original dataset. Color of the pixel describes its planarity and identifies the plane it belongs to. We use black color ([0, 0, 0]) as a mark of not planar pixels. Moreover, our annotations are associated between consecutive frames, so each plane has one and only one color which its pixels can have.

Annotation for LiDaR data is saved as an appropriately named numpy format file (.npy) where one file describes one cloud of the same name from the original dataset sequence. Annotation file contains an array of integers. Its size is equal to the amount of points in the respective point cloud and integer in the position i is plain id for point at the same position in the point cloud file. While 0 is used as an id for not planar points, other positive integers are used as unique identifiers of the plane along the whole sequence.