✨
AI Summary
- M-Attack: effective attacks against strong vision-language models.
- Refines semantic details with random cropping and alignment.
- Overcomes limitations of uniform and vague perturbations.
Neural intel Pod