Semantic segmentation on 3D point clouds is an important task for 3D sceneunderstanding . We train a semantic point cloud segmentation network with only asmall portion of points being labeled . We argue that we can better utilize thelimited supervision information as we densely propagate the supervision signalfrom the labeled points to other points within and across the input samples . We propose a cross-sample feature reallocating module to transfersimilar features and therefore re-route the gradients across two samples withcommon classes . Our weakly supervised method with only 10\% and 1\% of labels can producecompatible results with the fully supervised counterpart. We conduct extensive experiments on public datasets S3DIS and ScanNet. Ourweakly supervised methods can produce compatible results with a fully supervised version of our fully supervised counterparts. We propose to re-evaluate the

Author(s) : Jiacheng Wei, Guosheng Lin, Kim-Hui Yap, Fayao Liu, Tzu-Yi Hung

Links : PDF - Abstract

Code :
Coursera

Keywords : supervised - segmentation - d - point - supervision -

Leave a Reply

Your email address will not be published. Required fields are marked *