Vision-LLMs Can Fool Themselves with Self-Generated Typographic Attacks
Fuente:
arXiv
Enregistré dans:
| Auteurs principaux: | , , , , |
|---|---|
| Format: | Preprint |
| Publié: |
2024
|
| Sujets: | |
| Accès en ligne: | |
| Tags: |
Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
|
| _version_ | 1866913688804392960 |
|---|---|
| author | Qraitem, Maan Tasnim, Nazia Teterwak, Piotr Saenko, Kate Plummer, Bryan A. |
| author_facet | Qraitem, Maan Tasnim, Nazia Teterwak, Piotr Saenko, Kate Plummer, Bryan A. |
| contents | Typographic attacks, adding misleading text to images, can deceive vision-language models (LVLMs). The susceptibility of recent large LVLMs like GPT4-V to such attacks is understudied, raising concerns about amplified misinformation in personal assistant applications. Previous attacks use simple strategies, such as random misleading words, which don't fully exploit LVLMs' language reasoning abilities. We introduce an experimental setup for testing typographic attacks on LVLMs and propose two novel self-generated attacks: (1) Class-based attacks, where the model identifies a similar class to deceive itself, and (2) Reasoned attacks, where an advanced LVLM suggests an attack combining a deceiving class and description. Our experiments show these attacks significantly reduce classification performance by up to 60\% and are effective across different models, including InstructBLIP and MiniGPT4. Code: https://github.com/mqraitem/Self-Gen-Typo-Attack |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2402_00626 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | Vision-LLMs Can Fool Themselves with Self-Generated Typographic Attacks Qraitem, Maan Tasnim, Nazia Teterwak, Piotr Saenko, Kate Plummer, Bryan A. Computer Vision and Pattern Recognition Cryptography and Security Machine Learning Typographic attacks, adding misleading text to images, can deceive vision-language models (LVLMs). The susceptibility of recent large LVLMs like GPT4-V to such attacks is understudied, raising concerns about amplified misinformation in personal assistant applications. Previous attacks use simple strategies, such as random misleading words, which don't fully exploit LVLMs' language reasoning abilities. We introduce an experimental setup for testing typographic attacks on LVLMs and propose two novel self-generated attacks: (1) Class-based attacks, where the model identifies a similar class to deceive itself, and (2) Reasoned attacks, where an advanced LVLM suggests an attack combining a deceiving class and description. Our experiments show these attacks significantly reduce classification performance by up to 60\% and are effective across different models, including InstructBLIP and MiniGPT4. Code: https://github.com/mqraitem/Self-Gen-Typo-Attack |
| title | Vision-LLMs Can Fool Themselves with Self-Generated Typographic Attacks |
| topic | Computer Vision and Pattern Recognition Cryptography and Security Machine Learning |
| url | https://arxiv.org/abs/2402.00626 |