Nearest Gradient Calibration and Loss-Prioritized Gradient Decorrelation reduce the AP@0.5 of a camouflaged vehicle detector from 13.19% to 2.16% with YOLOv3, and improve transfer to other detectors.
Title resolution pending
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
-
Physical Adversarial Camouflage through Gradient Calibration and Regularization
Nearest Gradient Calibration and Loss-Prioritized Gradient Decorrelation reduce the AP@0.5 of a camouflaged vehicle detector from 13.19% to 2.16% with YOLOv3, and improve transfer to other detectors.