New method enhances adversarial attacks on semantic segmentation models

Researchers have developed IGME, an efficient method for generating transferable adversarial perturbations for semantic segmentation models. This approach uses a single source model to compose attack components, sharing gradient computations to reduce costs. IGME employs an integrated-gradient-style path-averaged direction to stabilize updates and demonstrates competitive transferability and runtime efficiency compared to existing methods on CNN- and transformer-based models. AI

IMPACT This research could lead to more robust defenses against adversarial attacks in computer vision applications.

RANK_REASON The item is an academic paper detailing a new method for adversarial attacks on computer vision models. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CV →

• Cityscapes
• CNN
• PASCAL-VOC
• semantic segmentation
• Transformer++

• paper
• other

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