ViP: Unified Certified Detection and Recovery for Patch Attack with Vision Transformers

Keyphrases

• ImageNet 100%
• Art Performance 100%
• Transformer 100%
• Adversarial Patch 100%
• Dual Patches 100%
• Patch Defence 100%
• Patch Attack 100%
• Unified Framework 50%
• Latest Progress 50%
• Closing the Gap 50%
• Attacker 50%
• Image Accuracy 50%
• Masked Language Model 50%
• Certified Robustness 50%
• Certified Accuracy 50%
• Masked Autoencoder 50%
• Self-Supervised Vision Transformers 50%
• Vision Transformer 50%

Computer Science

• Art Performance 100%
• Vision Transformer 100%
• Unified Framework 50%
• Image Modeling 50%
• Attackers 50%
• New-State 50%
• Autoencoder 50%