Reproduce “LiDAR-SHIELD: A Fully Dockerized Adversarial Robustness Assessment Framework for LiDAR-Based 3D Perception Models” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Alexandros Gkillas, Ioulia Kapsali, Christos Anagnostopoulos, Aris S. Lalos
LiDAR-SHIELD provides a framework for adversarial testing of LiDAR segmentation and detection. It separates dataset adapters and model services from an assessment engine that runs attacks through shared interfaces while models remain in their native environments. In black-box tests, LENet’s mIoU fell from 64.70 to 19.50, and SECOND’s mAP fell from 70.43 to 52.51. These results show vulnerability only in tested configurations.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.