New multi-view adversarial attack targets 3D vision models

Researchers have developed MVAP-G, a novel method for generating multi-view adversarial perturbations specifically designed to attack the Visual Geometry Grounded Transformer (VGGT). This new technique allows for the creation of consistent, imperceptible perturbations across multiple image views in a single feed-forward pass, overcoming the limitations of previous adversarial attack methods. Experiments show that MVAP-G significantly degrades VGGT's 3D reconstruction performance without requiring iterative optimization during inference, highlighting critical security vulnerabilities in 3D vision foundation models. AI

IMPACT Highlights potential security vulnerabilities in 3D vision foundation models, underscoring the need for more robust systems.

RANK_REASON Academic paper detailing a new method for adversarial attacks on a specific AI model. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CV →

• alphaXiv
• arXiv
• CatalyzeX
• DagsHub
• Gotit.pub
• Hugging Face
• MVAP-G
• ScienceCast
• VGGT
• Visual Geometry Grounded Transformer

• paper
• safety

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →