ETV-Attack: Efficient text-driven visual-variable adversarial attacks on visual question answering with pre-trained language models

Quanxing Xu

doi:10.1016/j.patcog.2026.113202

ETV-Attack: Efficient text-driven visual-variable adversarial attacks on visual question answering with pre-trained language models

2026-02-04

0

PRE

AI

DOI：10.1016/j.patcog.2026.113202

Original

Original request for help

Save

Abstract

En 中文

• This work provides an overall evaluation of LLM-based VQA in attack. • This work designs a text-driven embedding-based attack method. • This work proposes a text-driven augmentation module for VQA. • This work achieves new SOTA results on relevant benchmarks.

Keywords:

LLM-based VQA

text-driven attack

visual-variable adversarial attacks

pre-trained language models

VQA augmentation

ETV-Attack: Efficient text-driven visual-variable adversarial attacks on visual question answering with pre-trained language models

Abstract

Journal

Organization