A text-conditioned U-Net can generate input-aware triggers that backdoor VLM visual grounding, forcing the model to output the attacker-chosen object's bounding box regardless of the user query.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
IAG: Input-aware Backdoor Attack on VLM-based Visual Grounding
A text-conditioned U-Net can generate input-aware triggers that backdoor VLM visual grounding, forcing the model to output the attacker-chosen object's bounding box regardless of the user query.