Skip to content

Supplementary material for paper D. Bobkov et al. "Noise-resistant Unsupervised Object Segmentation in Multi-view Indoor Point Clouds", 2017. The paper is presented in 12th International Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Applications in Porto, Portugal in February 2017

License

Notifications You must be signed in to change notification settings

DBobkov/segmentation

Repository files navigation

Object segmentation in point cloud data

Supplementary material for paper D. Bobkov, S. Chen, M. Kiechle, S. Hilsenbeck, E. Steinbach "Noise-resistant Unsupervised Object Segmentation in Multi-view Indoor Point Clouds", 2017. The paper is presented in 12th International Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Applications in Porto, Portugal in February 2017.

This project contains:

  1. Code to perform multi-scale evaluation of segmentation on 3D data. For this you need a labelled result along with two files for ground truth. This is based on the segmentation evaluation metric of Richtsfeld et al. "Segmentation of unknown objects in indoor environments", IROS 2012. We extended it to include various possible object scales. For more details see the report file.

  2. Point cloud datasets with ground truth labelling on two levels. This includes 6 scenes within an office environment containing various objects, in total 452 object parts and 156 objects. For quick overview, proceed to web-viewer located here http://www.navvis.com/iv.hmw/.

  3. Raw pointcloud of the entire buildings (3 used buildings with XYZ and RGB information for each point) can be accessed here http://www.lmt.ei.tum.de/fileadmin/user_upload/bobkov/Object_segmentation_dataset/object_segmentation_building_dataset.zip

Explanation of the labelling

Coarse labelling (objects)

Coarse labelling

Fine labelling (object parts)

Fine labelling

Overview of scenes

Room1

Room1

Room2

Room2

Room3

Room2

Room4

Room2

Room5

Room2

Room6

Room2

About

Supplementary material for paper D. Bobkov et al. "Noise-resistant Unsupervised Object Segmentation in Multi-view Indoor Point Clouds", 2017. The paper is presented in 12th International Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Applications in Porto, Portugal in February 2017

Topics

Resources

License

Code of conduct

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published