Labeling datasets with OLT

The Object Labeling Toolkit (OLT) consists of a number of software applications for performing an effortless labeling of datasets, concretely those containing sequences of RGB-D observations.

OLT logo



If you use this software, please cite it through:

      @INPROCEEDINGS{Ruiz-Sarmiento-ECMR-2015,
           author = {Ruiz-Sarmiento, J. R. and Galindo, Cipriano and Gonz{\'{a}}lez-Jim{\'{e}}nez, Javier},
            title = {OLT: A Toolkit for Object Labeling Applied to Robotic RGB-D Datasets},
        booktitle = {European Conference on Mobile Robots},
             year = {2015},
         location = {Lincoln, UK}
      }
      

For getting the toolkit and further information please check the project webpage at GitHub. OLT aims at facilitate the labeling of arbitrarily large sequences of RGB-D observations. For that, the point clouds from the RGB-D observations are registered in order to reconstruct the captured scene. Then, the toolkit provides an application to easily annotate objects within that reconstruction by fitting boxes to them:




Example of a kitchen scene being annotated.


These annotations are finally propagated to each individual RGB-D observation in the sequence, resulting in a dense labeling of their RGB and depth information.

The main software components (appliations) of the toolkit are:

  • Process_rawlog: Sets the extrinsic and intrinsic parameters of the sensors used within the dataset.
  • Localization: Localizes the poses/locations from which the RGB-D observations of the datset were taken.
  • Sequential_visualization: Visually shows a 3D reconstruction of the collected data, and stores it as a scene.
  • Label_scene: Permits us to effortlessly label a reconstructed scene.
  • Label_rawlog: Propagates the annotated labels in a scene to each RGB-D observation (to both their RGB and depth information) within the dataset.
  • Dataset_statistics: Shows information of the dataset, e.g. a summary of the objects appearing on it, number of times that they appear, number of pixels they occupy, etc.
  • Create_video & Segmentation: Experimental applications under development.

The toolkit resorts to two widely used libraries in robotics: