A Real-World Dataset for Evaluating Physical Object Trigger Backdoor Attacks in the Wild
A Real-World Dataset for Evaluating Physical Object Trigger Backdoor Attacks in the Wild
Embedded Files
Detection Tasks
Detection Tasks
Traffic Sign Detection Task (Recorded from cars)
Vehicle Detection Task (Recorded from drones)
Data Collection & Dataset Descriptions
Data Collection & Dataset Descriptions
We employed four different triggers: Post-It Note, Flower, Target, and RGB.
We also included clean videos without triggers.
All of the 44 videos, over 32,000 frames, are summarised in Table 1.
The traffic sign detection dataset consists of 3840 × 2160 resolution videos captured by a dashboard-mounted Samsung Galaxy phone camera inside a car driving by different roadside traffic signs at various speeds (30-80 km/h) and from distances of 10-60 m.
The vehicle detection dataset consists of 1920 × 1440 resolution videos of a GoPro mounted on a drone flying at approximately 20 km/h at 20 m above driving cars for the vehicle detection task. These videos showcase various objects of interest, such as traffic signs and cars under diverse lighting conditions, at different times of the day, and from different distances and angles; these are significant factors that can impact the effectiveness of physical-world attacks.
Table 1: Detailed description of the 44 videos included in the DriveByFlyBy Dataset. A: Directly under sunlight without any shade. B: Lower sunlight conditions with shade from surrounding objects such as trees.
Structure of The Dataset
Structure of The Dataset
Under the physical_dataset, there are 8 subfolders named by the trigger type and position. All of videos related to that specific trigger type and position will be included in the subfolder.
blue_out: The Post-It Note trigger is attached outside of the traffic sign
blue_low: The Post-It Note trigger is attached at the bottom of the traffic sign
blue_high: The Post-It Note trigger is attached at the top of the traffic sign
blue_multi: The Post-It Note trigger is attached at the bottom of the traffic sign
flower: The flower sticker trigger is attached at the bottom of the traffic sign
target: The target sticker trigger is attached on top of the car
rgb: The RGB sticker trigger is attached on top of the car
All datasets and videos on this page and linked to from this page are copyright by us and published under the Creative Commons Attribution-NonCommercial-ShareAlike 3.0 License. This means that you must attribute the work in the manner specified by the authors, you may not use this work for commercial purposes and if you alter, transform, or build upon this work, you may distribute the resulting work only under the same license.
If you use this dataset, please cite us as follows
Bao Gia Doan, Dang Quang Nguyen, Callum Lindquist, Paul Montague, Tamas Abraham, Olivier De Vel, Seyit Camtepe, Salil S. Kanhere, Ehsan Abbasnejad, and Damith C. Ranasinghe. On the Credibility of Backdoor Attacks Against Object Detectors in the Physical World. In Annual Computer Security Applications Conference (ACSAC), 2024.
Download The Dataset
Download The Dataset
Privacy
Privacy
This dataset is made available for academic use only. However, we take your privacy seriously! If you find yourself or personal items in this dataset and feel unhappy about their use, please contact us via this email (drivebyflyby@gmail.com) and we will immediately remove the respective data from the dataset.
Page updated
Google Sites
Report abuse