Bosch Semantic Interpretation Challenge (Indoor) === Introduction === The Bosch Semantic Interpretation Challenge (indoor) dataset contains data of 10 different indoor scenes (apartments). With this data set we want to help researchers in the domain of mobile robotics test their algorithms for semantic segmentation. For each scene, we provide: - high precision and downsampled colored point clouds - a high precision 3D Mesh - views rendered from several pre-computed viewpoints in the scene - 2D projection ground truth data for geometric segmentation (rooms defined by the building's structure) - 2D projection ground truth data for semantic segmentation (rooms are defined by geometry and semantic use; e.g. a single physical room might be further separated, if there's a dining area and a kitchen area) === Data Acquisition === This work was conducted at the Bosch North America Research department, Palo Alto, California. The viewpoints for rendering views were generated as described in [1]: === Content === The dataset is structured as follow: - For each scene, there is a scene_name.tar.gz file. After extracting a scene you'll find the following files and folders: Folders: - mesh: Contains a renderable, textured, mesh of the scene. The format is Wavefront .obj. - mesh_views: Contains views rendered from the viewpoints given in scene_name/viewpoints.txt. The views are generated from a position halfway between the floor and the ceiling. - rooms_categorization: - ground_truth.png: ground truth 2d segmentation of the room (color coded see below) - ground_truth.txt: the projection parameters. - room_X.pcd: The points belonging to a room - room_X.txt: Label of a room Files: - scene_name.pcd: a lower quality pointcloud - scene_name_high.pcd: (not available in sunnyvale) high quality pointcloud - viewpoint_cloud.pcd: Viewpoints used for rendering in pcd format. - viewpoints.txt: Viewpoints used for rendering in text format. === Projection === The projection from a 3D point to the 2D map is done by the following cpp snipped: ===================================== float mesh_2d_projection_resolution; cv::Point2d mesh_2d_projection_origin_shift; std::ifstream in("ground_truth.txt"); in >> mesh_2d_projection_resolution >> mesh_2d_projection_origin_shift.x >> mesh_2d_projection_origin_shift.y; cv::Point2i pos_2d( (pos(0) + mesh_2d_projection_origin_shift.x) / mesh_2d_projection_resolution, (pos(1) + mesh_2d_projection_origin_shift.y) / mesh_2d_projection_resolution); ==================================== The parameters can be found in rooms_categorization/ground_truth.txt === Color coding and semantic labels === In the semantig gound truth 2D projections, the room types are represented by the following colors: label_code["bathroom"] = cv::Vec3b(255,0,0); label_code["bedroom"] = cv::Vec3b(0,255,0); label_code["childs_room"] = cv::Vec3b(0,0,255); label_code["corridor"] = cv::Vec3b(255,255,0); label_code["dining_room"] = cv::Vec3b(0,255,255); label_code["kitchen"] = cv::Vec3b(255,153,204); label_code["laundromat"] = cv::Vec3b(204,255,204); label_code["living_room"] = cv::Vec3b(51,0,0); label_code["office"] = cv::Vec3b(0,128,255); label_code["staircase"] = cv::Vec3b(51,102,0); label_code["storage_room"] = cv::Vec3b(102,102,255); label_code["unknown"] = cv::Vec3b(0,0,0); === Publications === This is a collection of publications using the "Bosch Semantic Interpretation Challenge (Indoor)" dataset. If you used the dataset for your own publication, please contact us to get listed here. [1] R. Ambruş, S. Claici, and A. Wendt, “Automatic room segmentation from unstructured 3-d data of indoor environments,” IEEE Robotics and Automation Letters, vol. 2, no. 2, pp. 749–756, April 2017 [2] M. Durner*, M. Brucker*, R. Ambrus*, Z.-C. Marton, A. J. Wendt, P. Jensfelt, K. O. Arras, R. Triebel. Semantic Labeling of Indoor Environments from 3D RGB Maps. In 2017 IEEE International Conference on Robotics and Automation (Late Breaking Results Poster Session). === Download === The dataset can be downloaded at: http://rmc.dlr.de/download/bosch_semantic_interpretation_challenge_indoor