SHIFT3D: Synthesizing Hard Inputs For Tricking 3D Detectors
Not on the map. Why?
Abstract (arXiv)
We present SHIFT3D, a differentiable pipeline for generating 3D shapes that are structurally plausible yet challenging to 3D object detectors. In safety-critical applications like autonomous driving, discovering such novel challenging objects can offer insight into unknown vulnerabilities of 3D detectors. By representing objects with a signed distanced function (SDF), we show that gradient error signals allow us to smoothly deform the shape or pose of a 3D object in order to confuse a downstream 3D detector. Importantly, the objects generated by SHIFT3D physically differ from the baseline object yet retain a semantically recognizable shape. Our approach provides interpretable failure modes for modern 3D object detectors, and can aid in preemptive discovery of potential safety risks within 3D perception systems before these risks become critical failures.
Key facts. Cited by 1 paper in the map (data through 7 Oct 2026). Builds on 30 of its 33 references that are in the map. First posted on arXiv 11 Sep 2023.
Builds on
30 of 33 references.
References · 30
- 2014Explaining and Harnessing Adversarial Examplescited by 7,287
- 2013Intriguing properties of neural networkscited by 5,114
- 2015ShapeNet: An Information-Rich 3D Model Repositorycited by 2,880
- 2019DeepSDF: Learning Continuous Signed Distance Functions for Shape Representationcited by 2,682
- 2019Scalability in Perception for Autonomous Driving: Waymo Open Datasetcited by 2,436
- 2018PointPillars: Fast Encoders for Object Detection from Point Cloudscited by 1,707
- 2015DeepFool: a simple and accurate method to fool deep neural networkscited by 1,315
- 2019Large-Scale Long-Tailed Recognition in an Open Worldcited by 704
- 2017Adversarial Examples for Semantic Segmentation and Object Detectioncited by 275
- 2018Generating Adversarial Examples with Adversarial Networkscited by 256
- 2021Embracing Single Stride 3D Object Detector with Sparse Transformercited by 193
- 2018Generating 3D Adversarial Point Cloudscited by 144
18 more
- 2018Strike (with) a Pose: Neural Networks Are Easily Fooled by Strange Poses of Familiar Objectscited by 143
- 2020Physically Realizable Adversarial Examples for LiDAR Object Detectioncited by 137
- 2018Constructing Unrestricted Adversarial Examples with Generative Modelscited by 135
- 2018PointCloud Saliency Mapscited by 112
- 2021Invisible for both Camera and LiDAR: Security of Multi-Sensor Fusion based Perception in Autonomous Driving Under Physical-World Attackscited by 101
- 2019Extending Adversarial Attacks and Defenses to Deep 3D Point Cloud Classifierscited by 92
- 2017Adversarial Attacks Beyond the Image Spacecited by 67
- 2019Adversarial Attack and Defense on Point Setscited by 63
- 2019Adversarial Objects Against LiDAR-Based Autonomous Driving Systemscited by 62
- 2019Semantic Adversarial Attacks: Parametric Transformations That Fool Deep Classifierscited by 56
- 2019Robustness of 3D Deep Learning in an Adversarial Settingcited by 55
- 2020LG-GAN: Label Guided Adversarial Network for Flexible Targeted Attack of Point Cloud-based Deep Networkscited by 55
- 2021ReDAL: Region-based and Diversity-aware Active Learning for Point Cloud Semantic Segmentationcited by 52
- 2020Minimal Adversarial Examples for Deep Learning on 3D Point Cloudscited by 41
- 2021Active Learning at the ImageNet Scalecited by 28
- 2020SAMP: Shape and Motion Priors for 4D Vehicle Reconstructioncited by 19
- 2022Improving the Intra-class Long-tail in 3D Detection via Rare Example Miningcited by 19
- 2022GradTail: Learning Long-Tailed Data Using Gradient-based Sample Weightingcited by 4
Built on by
Citing papers per year: 2024: 1.
1 paper.