LLM-generated adversarial captions, selected and self-trained with diversity-promoting filtering, deceive CLIP-style multimodal representations at high rates across image, video, and audio.
Generated Caption:
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Can LLMs Deceive CLIP? Benchmarking Adversarial Compositionality of Pre-trained Multimodal Representation via Text Updates
LLM-generated adversarial captions, selected and self-trained with diversity-promoting filtering, deceive CLIP-style multimodal representations at high rates across image, video, and audio.