Forensics-Bench
收藏资源简介:
Forensics-Bench是由香港大学等研究机构提出的一个新的伪造检测评估基准套件,旨在全面评估大型视觉语言模型(LVLMs)在伪造检测方面的能力。该数据集包含63292个精心挑选的多选视觉问题,涵盖了112种独特的伪造检测类型,从五个不同的角度(伪造语义、伪造模态、伪造任务、伪造类型和伪造模型)对伪造进行描述。数据集的内容包括RGB图像、近红外图像、视频和文本等多种模态,覆盖了人像和一般主题的多种语义,并包含了由不同AI模型创建或操作的各种伪造类型。该数据集的设计旨在推动LVLMs在伪造检测领域的发展,解决AI生成内容带来的挑战。
Forensics-Bench is a novel forgery detection evaluation benchmark suite proposed by research institutions including the University of Hong Kong, aiming to comprehensively evaluate the forgery detection capabilities of Large Vision-Language Models (LVLMs). This dataset comprises 63,292 carefully curated multiple-choice visual questions, covering 112 unique forgery detection categories, and characterizes forgeries from five distinct perspectives: forgery semantics, forgery modalities, forgery tasks, forgery types, and forgery models. The dataset includes diverse modalities such as RGB images, near-infrared images, videos and text, covers diverse semantics of portraits and general topics, and encompasses various forgery types created or manipulated by different AI models. This benchmark is designed to advance the development of LVLMs in the field of forgery detection and address the challenges posed by AI-generated content.




