Researchers have identified a new type of physical adversarial attack on vision systems, termed SLASH (Scratch-induced Lens Adversarial Streak Hijacking). This attack exploits small scratches on camera lenses or protective covers, which, when interacting with light sources, create structured streak artifacts that distort depth cues. The attack is persistent as the damage is fixed, but selective as it is triggered by specific scene conditions, leading to significant errors in monocular depth estimation and 3D object detection. AI
IMPACT Reveals a new attack surface where physical imperfections can be exploited to compromise vision system accuracy, necessitating new defenses.
RANK_REASON The cluster contains a research paper detailing a novel attack method on computer vision systems.
- arXiv
- cs.CV
- monocular 3D object detection
- monocular depth estimation
- Scratched Lenses, Shifted Depth: Passive Camera-Side Optical Attacks
- SLASH
AI-generated summary · Google Gemini · from 2 sources. How we write summaries →