Home
Uni-Logo
 

Orientation-boosted Voxel Nets for 3D Object Recognition

Technical Report , arXiv:1604.03351, 2016
Download the publication : 1604.03351v1.pdf [5.6Mo]  

Abstract: Recent work has shown good recognition results in 3D data using 3D convolutional networks. In this paper, we argue that the object orientation plays an important role in 3D recognition. To this end, we approach the category-level classification task as a multi-task problem, in which the network is forced to predict the pose of the object in addition to the class label. We show that this yields significant improvements in the classification results. We implemented different network architectures for this purpose and tested them on different datasets representing various 3D data sources: LiDAR data, CAD models and RGBD images. We report state-of-the-art results on classification, and analyze the effects of orientation-boosting on the dominant signal paths in the network.

Images and movies

 

See also

BibTex references

@TechReport{SZB16,
  author       = "N. Sedaghat and M. Zolfaghari and T. Brox",
  title        = "Orientation-boosted Voxel Nets for 3D Object Recognition",
  institution  = "arXiv:1604.03351",
  year         = "2016",
  url          = "http://lmb.informatik.uni-freiburg.de//Publications/2016/SZB16"
}

Other publications in the database