IAG: Input-aware Backdoor Attack on VLMs for Visual Grounding
Research paper by Junxian Li and collaborators. Published 2025-08-13. Topics: vision-language models, visual grounding, backdoor attacks, input-aware backdoor attack, adaptive trigger generator.
Resource information
- Authors
- Junxian Li, Beining Xu, Di Zhang
- Research topics
- vision-language models, visual grounding, backdoor attacks, input-aware backdoor attack, adaptive trigger generator, text-conditional U-Net, reconstruction loss, ASR@0.5, InternVL-2.5-8B, Ferret-7B, LlaVA-1.5-7B, ablation study
- Publication date
- 2025-08-13
- Publication
- arXiv
- Last record update
- Nov 3, 2026