Revisiting Point Cloud Shape Classification with a Simple and Effective Baseline

Ankit Goyal, Hei Law, Bowei Liu, Alejandro Newell, Jia Deng
Proceedings of the 38th International Conference on Machine Learning, PMLR 139:3809-3820, 2021.

Abstract

Processing point cloud data is an important component of many real-world systems. As such, a wide variety of point-based approaches have been proposed, reporting steady benchmark improvements over time. We study the key ingredients of this progress and uncover two critical results. First, we find that auxiliary factors like different evaluation schemes, data augmentation strategies, and loss functions, which are independent of the model architecture, make a large difference in performance. The differences are large enough that they obscure the effect of architecture. When these factors are controlled for, PointNet++, a relatively older network, performs competitively with recent methods. Second, a very simple projection-based method, which we refer to as SimpleView, performs surprisingly well. It achieves on par or better results than sophisticated state-of-the-art methods on ModelNet40 while being half the size of PointNet++. It also outperforms state-of-the-art methods on ScanObjectNN, a real-world point cloud benchmark, and demonstrates better cross-dataset generalization. Code is available at https://github.com/princeton-vl/SimpleView.

Cite this Paper

BibTeX
@InProceedings{pmlr-v139-goyal21a, title = {Revisiting Point Cloud Shape Classification with a Simple and Effective Baseline}, author = {Goyal, Ankit and Law, Hei and Liu, Bowei and Newell, Alejandro and Deng, Jia}, booktitle = {Proceedings of the 38th International Conference on Machine Learning}, pages = {3809--3820}, year = {2021}, editor = {Meila, Marina and Zhang, Tong}, volume = {139}, series = {Proceedings of Machine Learning Research}, month = {18--24 Jul}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v139/goyal21a/goyal21a.pdf}, url = {https://proceedings.mlr.press/v139/goyal21a.html}, abstract = {Processing point cloud data is an important component of many real-world systems. As such, a wide variety of point-based approaches have been proposed, reporting steady benchmark improvements over time. We study the key ingredients of this progress and uncover two critical results. First, we find that auxiliary factors like different evaluation schemes, data augmentation strategies, and loss functions, which are independent of the model architecture, make a large difference in performance. The differences are large enough that they obscure the effect of architecture. When these factors are controlled for, PointNet++, a relatively older network, performs competitively with recent methods. Second, a very simple projection-based method, which we refer to as SimpleView, performs surprisingly well. It achieves on par or better results than sophisticated state-of-the-art methods on ModelNet40 while being half the size of PointNet++. It also outperforms state-of-the-art methods on ScanObjectNN, a real-world point cloud benchmark, and demonstrates better cross-dataset generalization. Code is available at https://github.com/princeton-vl/SimpleView.} }
Endnote
%0 Conference Paper %T Revisiting Point Cloud Shape Classification with a Simple and Effective Baseline %A Ankit Goyal %A Hei Law %A Bowei Liu %A Alejandro Newell %A Jia Deng %B Proceedings of the 38th International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2021 %E Marina Meila %E Tong Zhang %F pmlr-v139-goyal21a %I PMLR %P 3809--3820 %U https://proceedings.mlr.press/v139/goyal21a.html %V 139 %X Processing point cloud data is an important component of many real-world systems. As such, a wide variety of point-based approaches have been proposed, reporting steady benchmark improvements over time. We study the key ingredients of this progress and uncover two critical results. First, we find that auxiliary factors like different evaluation schemes, data augmentation strategies, and loss functions, which are independent of the model architecture, make a large difference in performance. The differences are large enough that they obscure the effect of architecture. When these factors are controlled for, PointNet++, a relatively older network, performs competitively with recent methods. Second, a very simple projection-based method, which we refer to as SimpleView, performs surprisingly well. It achieves on par or better results than sophisticated state-of-the-art methods on ModelNet40 while being half the size of PointNet++. It also outperforms state-of-the-art methods on ScanObjectNN, a real-world point cloud benchmark, and demonstrates better cross-dataset generalization. Code is available at https://github.com/princeton-vl/SimpleView.
APA
Goyal, A., Law, H., Liu, B., Newell, A. & Deng, J.. (2021). Revisiting Point Cloud Shape Classification with a Simple and Effective Baseline. Proceedings of the 38th International Conference on Machine Learning, in Proceedings of Machine Learning Research 139:3809-3820 Available from https://proceedings.mlr.press/v139/goyal21a.html.

Related Material