Flex-Convolution

Million-Scale Point-Cloud Learning Beyond Grid-Worlds

Fabian Groh, Patrick Wieschollek and Hendrik P. A. Lensch
University of Tübingen
Asian Conference on Computer Vision (ACCV) 2018

Hier wäre eigentlich ein Video zu sehen. Damit Sie diesen Inhalt (Quelle: www.xyz.de) sehen können, klicken Sie bitte auf "Akzeptieren". Wir weisen Sie darauf hin, dass durch das Abspielen des Videos neben Cookies zur Videowiedergabe auch Cookies des Drittanbieters zu Targeting- und Werbezwecken gesetzt und Informationen ggf. mit weiteren Diensten des Drittanbieters verknüpft werden können. Weitere Informationen und eine Widerrufsmöglichkeit finden Sie in unserer Datenschutzerklärung.

Abstract

Traditional convolution layers are specifically designed to exploit the natural data representation of images -- a fixed and regular grid. However, unstructured data like 3D point clouds containing irregular neighborhoods constantly breaks the grid-based data assumption. Therefore applying best-practices and design choices from 2D-image learning methods towards processing point clouds are not readily possible. In this work, we introduce a natural generalization flex-convolution of the conventional convolution layer along with an efficient GPU implementation. We demonstrate competitive performance on rather small benchmark sets using fewer parameters and lower memory consumption and obtain significant improvements on a million-scale real-world dataset. Ours is the first which allows to efficiently process 7 million points concurrently.

Content

Our network can handle up to handle inputs up to 7 on a NVIDIA 1080Ti in a single forward pass with the same accuracy.

It scales beyond, up to 17 Million points on a V100 GPU.

It therefore processes entire rooms at full-resolution without the need of sub-sampling.

Semantic point cloud segmentation produced as raw outputs of our proposed network from the held-out validation set. In this point-based rendering, surfaces might not be illustrated as opaque. Interestingly, in room C our network classifies the beanbag as “sofa”, while its ground-truth annotation is “chair”. For more results please refer to the accompanying video.

More Resources

Paper

Code

Bibtex

@inproceedings{accv2018/Groh,
author = {Fabian Groh and Patrick Wieschollek and Hendrik P. A. Lensch },
  title = {Flex-Convolution (Million-Scale Pointcloud Learning Beyond Grid-Worlds)},
  booktitle = {Asian Conference on Computer Vision (ACCV)},
  month = {Dezember},
  year = {2018}
}

Million-Scale Point-Cloud Learning Beyond Grid-Worlds

Externer Inhalt