%0 Journal Article %1 you2018pointwise %A You, Yang %A Lou, Yujing %A Liu, Qi %A Ma, Lizhuang %A Wang, Weiming %A Tai, Yuwing %A Lu, Cewu %D 2018 %K equivariance partsegmentation pointcloud rotationinvariant semanticsegmentation shapenet so3 sphericalcnns sphericalharmonics %T PRIN: Pointwise Rotation-Invariant Network %U http://arxiv.org/abs/1811.09361 %X In recent years, point clouds have earned quite some research interests by the development of depth sensors. Due to different layouts of objects, orientation of point clouds is often unknown in real applications. In this paper, we propose a new point sets learning framework named Pointwise Rotation-Invariant Network (PRIN), focusing on the rotation problem in point clouds. We construct spherical signals by adaptive sampling from sparse points and employ spherical convolutions, together with tri-linear interpolation to extract rotation-invariant features for each point. Our network can be applied in applications ranging from object classification, part segmentation, to 3D feature matching and label alignment. PRIN shows similar performance on par or better than state-of-the-art methods on part segmentation without data augmentation. We provide theoretical analysis for what our network has learned and why it is robust to input rotation. Our code is available online.

@article{you2018pointwise, abstract = {In recent years, point clouds have earned quite some research interests by the development of depth sensors. Due to different layouts of objects, orientation of point clouds is often unknown in real applications. In this paper, we propose a new point sets learning framework named Pointwise Rotation-Invariant Network (PRIN), focusing on the rotation problem in point clouds. We construct spherical signals by adaptive sampling from sparse points and employ spherical convolutions, together with tri-linear interpolation to extract rotation-invariant features for each point. Our network can be applied in applications ranging from object classification, part segmentation, to 3D feature matching and label alignment. PRIN shows similar performance on par or better than state-of-the-art methods on part segmentation without data augmentation. We provide theoretical analysis for what our network has learned and why it is robust to input rotation. Our code is available online.}, added-at = {2018-12-18T18:00:46.000+0100}, author = {You, Yang and Lou, Yujing and Liu, Qi and Ma, Lizhuang and Wang, Weiming and Tai, Yuwing and Lu, Cewu}, biburl = {https://www.bibsonomy.org/bibtex/29b2f74c283ef82f61c75afaa522ef387/rspezialetti}, description = {[1811.09361v3] PRIN: Pointwise Rotation-Invariant Network}, interhash = {3c838133e76c377b70551fa7944a2afd}, intrahash = {9b2f74c283ef82f61c75afaa522ef387}, keywords = {equivariance partsegmentation pointcloud rotationinvariant semanticsegmentation shapenet so3 sphericalcnns sphericalharmonics}, note = {cite arxiv:1811.09361}, timestamp = {2018-12-18T18:00:46.000+0100}, title = {PRIN: Pointwise Rotation-Invariant Network}, url = {http://arxiv.org/abs/1811.09361}, year = 2018 }