Benchmarking and Improving Robustness of 3D Point Cloud Recognition against Common Corruptions

Jiachen Sun, Qingzhao Zhang, Bhavya Kailkhura, Zhiding Yu, Chaowei Xiao, Zhuoqing Mao

Keywords: Corruption Robustness Benchmark, Point Cloud Classification, Data Augmentation

Abstract: Deep neural networks on 3D point cloud data have been widely used in the real world, especially in safety-critical applications. However, their robustness against corruptions is less studied. In this paper, we present ModelNet40-C, a comprehensive benchmark on 3D point cloud corruption robustness, consisting of 15 common and realistic corruptions. Our evaluation shows a significant gap between the performances on ModelNet40 and ModelNet40-C for state-of-the-art models. We identify a number of critical insights for future studies on corruption robustness in point cloud recognition. For instance, we unveil that Transformer-based architectures with proper training recipes achieve the strongest robustness. To bridge this gap, we further propose RobustNet and PointCutMixup that embrace the merits of existing architectural designs to further improve the corruption robustness in the 3D point cloud domain, after evaluating a wide range of augmentation and test-time adaptation strategies. Our codebase and dataset are open-sourced.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics

Submission Guidelines: Yes

Please Choose The Closest Area That Your Submission Falls Into: Infrastructure (eg, datasets, competitions, implementations, libraries)

TL;DR: We propose ModelNet40-C, a novel corruption robustness dataset and benchmark for point cloud recognition with RobustNet and PointCutMixup to further improve the rosbustness.