Publication
IAG: Input-Aware Backdoor Attack on VLM-Based Visual Grounding
Vision Language Models, Security
Cite this work View paper
Citation
Junxian Li, Beining Xu, Simin Chen, Jiatong Li, Jingdi Lei, Haodong Zhao, and Di Zhang. (2026). IAG: Input-Aware Backdoor Attack on VLM-Based Visual Grounding. CVPR 2026.
About this work
IAG studies input-aware backdoor attacks on vision-language models used for visual grounding. A text-conditioned UNet produces triggers associated with a specified target object description.
The evaluation covers models including LLaVA, InternVL, and Ferret, and benchmarks including RefCOCO, RefCOCO+, RefCOCOg, Flickr30k Entities, and ShowUI. It measures attack success, clean-input accuracy, and transfer across models and datasets.