A universal adversarial patch placed below a traffic light can flip red detections to green on YOLO models, with partial real-world success on a full autonomous driving perception pipeline.
Robust physical-world attacks on deep learning visual classification,
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Fool the Stoplight: Realistic Adversarial Patch Attacks on Traffic Light Detectors
A universal adversarial patch placed below a traffic light can flip red detections to green on YOLO models, with partial real-world success on a full autonomous driving perception pipeline.