Tracklet-Switch Adversarial Attack against Pedestrian Multi-Object Tracking Trackers
Not on the map. Why?
Abstract (arXiv)
Multi-Object Tracking (MOT) has achieved aggressive progress and derived many excellent deep learning trackers. Meanwhile, most deep learning models are known to be vulnerable to adversarial examples that are crafted with small perturbations but could mislead the model prediction. In this work, we observe that the robustness on the MOT trackers is rarely studied, and it is challenging to attack the MOT system since its mature association algorithms are designed to be robust against errors during the tracking. To this end, we analyze the vulnerability of popular MOT trackers and propose a novel adversarial attack method called Tracklet-Switch (TraSw) against the complete tracking pipeline of MOT. The proposed TraSw can fool the advanced deep pedestrian trackers (i.e., FairMOT and ByteTrack), causing them fail to track the targets in the subsequent frames by perturbing very few frames. Experiments on the MOT-Challenge datasets (i.e., 2DMOT15, MOT17, and MOT20) show that TraSw can achieve an extraordinarily high success attack rate of over 95% by attacking only four frames on average. To our knowledge, this is the first work on the adversarial attack against the pedestrian MOT trackers. Code is available at https://github.com/JHL-HUST/TraSw .
Key facts. Cited by 1 paper in the map (data through 14 Oct 2026). Builds on 52 of its 73 references that are in the map. First posted on arXiv 17 Nov 2021.
Builds on
52 of 73 references.
References · 52
- 2014Explaining and Harnessing Adversarial Examplescited by 7,364
- 2017Towards Deep Learning Models Resistant to Adversarial Attackscited by 6,415
- 2013Intriguing properties of neural networkscited by 5,155
- 2016Towards Evaluating the Robustness of Neural Networkscited by 2,998
- 2015FaceNet: A Unified Embedding for Face Recognition and Clusteringcited by 2,849
- 2016Adversarial examples in the physical worldcited by 1,769
- 2017Simple Online and Realtime Tracking with a Deep Association Metriccited by 910
- 2016Practical Black-Box Attacks against Machine Learningcited by 903
- 2016Simple Online and Realtime Trackingcited by 868
- 2021ByteTrack: Multi-Object Tracking by Associating Every Detection Boxcited by 823
- 2016Universal adversarial perturbationscited by 784
- 2014Deep Neural Networks are Easily Fooled: High Confidence Predictions for Unrecognizable Imagescited by 776
38 more
- 2016Performance Measures and a Data Set for Multi-Target, Multi-Camera Trackingcited by 692
- 2019CenterNet: Keypoint Triplets for Object Detectioncited by 654
- 2016MOT16: A Benchmark for Multi-Object Trackingcited by 532
- 2017Adversarial Patchcited by 517
- 2020FairMOT: On the Fairness of Detection and Re-Identification in Multiple Object Trackingcited by 490
- 2020Tracking Objects as Pointscited by 483
- 2018SiamRPN++: Evolution of Siamese Visual Tracking with Very Deep Networkscited by 446
- 2020MOT20: A benchmark for multi object tracking in crowded scenescited by 357
- 2019Nesterov Accelerated Gradient and Scale Invariance for Adversarial Attackscited by 355
- 2017Defense against Adversarial Attacks Using High-Level Representation Guided Denoisercited by 329
- 2019Towards Real-Time Multi-Object Trackingcited by 311
- 2017Adversarial Examples for Semantic Segmentation and Object Detectioncited by 275
- 2019Simple Black-box Adversarial Attackscited by 251
- 2014Multiple Object Tracking: A Literature Reviewcited by 166
- 2018ShapeShifter: Robust Physical Adversarial Attack on Faster R-CNN Object Detectorcited by 159
- 2019Deeper and Wider Siamese Networks for Real-Time Visual Trackingcited by 156
- 2018SoccerNet: A Scalable Dataset for Action Spotting in Soccer Videoscited by 150
- 2015MOTChallenge 2015: Towards a Benchmark for Multi-Target Trackingcited by 133
- 2020Chained-Tracker: Chaining Paired Attentive Regression Results for End-to-End Joint Multiple-Object Detection and Trackingcited by 115
- 2020Rethinking the competition between detection and ReID in Multi-Object Trackingcited by 101
- 2021TransMOT: Spatial-Temporal Graph Transformer for Multiple Object Trackingcited by 93
- 2017Universal Adversarial Perturbations Against Semantic Image Segmentationcited by 92
- 2018Features for Multi-Target Multi-Camera Tracking and Re-Identificationcited by 88
- 2016POI: Multiple Object Tracking with High Performance Detection and Appearance Featurecited by 74
- 2021Multiple Object Tracking with Correlation Learningcited by 67
- 2021SiamMOT: Siamese Multi-Object Trackingcited by 62
- 2017Adversarial Examples that Fool Detectorscited by 40
- 2017Adversarial Examples for Semantic Image Segmentationcited by 33
- 2020Cooling-Shrinking Attack: Blinding the Tracker with Imperceptible Noisescited by 32
- 2020Learning Black-Box Attackers with Transferable Priors and Query Feedbackcited by 32
- 2021Track without Appearance: Learn Box and Tracklet Embedding with Local and Global Motion Patterns for Vehicle Trackingcited by 26
- 2021IoU Attack: Towards Temporally Coherent Black-Box Adversarial Attack for Visual Object Trackingcited by 25
- 2017Regularizing deep networks using efficient layerwise adversarial trainingcited by 24
- 2017Biologically inspired protection of deep networks from adversarial attackscited by 23
- 2013A Survey of Appearance Models in Visual Object Trackingcited by 21
- 2017Multi-Target, Multi-Camera Tracking by Hierarchical Clustering: Recent Progress on DukeMTMC Projectcited by 19
- 2018Max-Mahalanobis Linear Discriminant Analysis Networkscited by 18
- 2019Fooling Detection Alone is Not Enough: First Adversarial Attack against Multiple Object Trackingcited by 9
50 most cited of 52 All 52 (JSON)
Built on by
Citing papers per year: 2026: 1.
1 paper.