arXiv cs.LGPaper
ARMOR: Manifold-Oriented Training for Adversarially Robust Aerial Object Detection under Data Scarcity
Competent work on a real problem (adversarial patches on drones in low-data settings), but it's narrow and domain-specific. The manifold-oriented approach is sound, but if you're not building aerial systems under data constraints, this is academic. The method itself isn't novel enough to generalize.