SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2402.00626 · 2024

Vision-LLMs Can Fool Themselves with Self-Generated Typographic Attacks

arXiv · PDF · Open in the Atlas

Code that ran

We lifted 8 functions out of this paper's own repositories and ran 4 of them in a sandbox. "Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results.

RepositoryRoleRan
mqraitem/self-gen-typo-attack canonical 4 of 8
FunctionStatusWhere it lives
encode_image Ran mqraitem/self-gen-typo-attack/utils_models/utils_gpt4.py
pointer only (licence: NONE) · get_code("f41cb1a19b154297")
get_gpt4_data Ran mqraitem/self-gen-typo-attack/utils.py
pointer only (licence: NONE) · get_code("95bcb0c3cf069e22")
get_gpt4_data_llm Ran mqraitem/self-gen-typo-attack/utils.py
pointer only (licence: NONE) · get_code("b9d472b7f2267e67")
get_random_data Ran mqraitem/self-gen-typo-attack/utils.py
pointer only (licence: NONE) · get_code("1c1ad9aaa08f3a38")
get_gpt4_llm_pred Not yet run mqraitem/self-gen-typo-attack/utils_models/utils_gpt4.py
pointer only (licence: NONE) · get_code("fb9b530e93968dac")
get_gpt4_pred Not yet run mqraitem/self-gen-typo-attack/utils_models/utils_gpt4.py
pointer only (licence: NONE) · get_code("e7615cd209323da1")
run_blip Not yet run mqraitem/self-gen-typo-attack/utils_models/utils_instructblip.py
pointer only (licence: NONE) · get_code("2e08e894022db180")
run_llava Not yet run mqraitem/self-gen-typo-attack/utils_models/utils_llava.py
pointer only (licence: NONE) · get_code("b607dd5d1f16b79b")

Repositories linked to this paper

Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.

Abstract

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

For agents

The same record, over MCP at https://syntology.ai/mcp:

get_harvested_code_for_paper("2402.00626")
get_code_for_paper("2402.00626")
have("2402.00626")

Connect an agent — have() is free.