POS Tag Adversarial Prompt Dataset
收藏arXiv2025-09-30 收录
下载链接：
https://github.com/shahariar-shibli/Adversarial-Attack-on-POS-Tags
下载链接
链接失效反馈官方服务：
资源简介：
该数据集是为了评估文本提示在文本到图像生成过程中，针对不同词性标签的对抗攻击而构建的。它包含了六个词性标签的输入和目标提示对，这六个词性标签分别是名词、动词、副词、形容词、数词和专有名词。该数据集规模包括118,287行标题和600个生成的提示对，其任务是对文本提示的对抗攻击评估。
This dataset is constructed to evaluate adversarial attacks targeting different part-of-speech (POS) tags in text prompts during text-to-image generation. It includes input-target prompt pairs across six part-of-speech categories: noun, verb, adverb, adjective, numeral, and proper noun. The dataset contains 118,287 caption lines and 600 generated prompt pairs, with its core task being adversarial attack evaluation for text prompts.
提供机构：
Authors of the paper
