eduzhai > Applied Sciences > Engineering >

Learning Graph-Convolutional Representations for Point Cloud Denoising

  • king
  • (0) Download
  • 20210505
  • Save

... pages left unread,continue reading

Document pages: 16 pages

Abstract: Point clouds are an increasingly relevant data type but they are oftencorrupted by noise. We propose a deep neural network based ongraph-convolutional layers that can elegantly deal with thepermutation-invariance problem encountered by learning-based point cloudprocessing methods. The network is fully-convolutional and can build complexhierarchies of features by dynamically constructing neighborhood graphs fromsimilarity among the high-dimensional feature representations of the points.When coupled with a loss promoting proximity to the ideal surface, the proposedapproach significantly outperforms state-of-the-art methods on a variety ofmetrics. In particular, it is able to improve in terms of Chamfer measure andof quality of the surface normals that can be estimated from the denoised data.We also show that it is especially robust both at high noise levels and inpresence of structured noise such as the one encountered in real LiDAR scans.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×