DiQuID
收藏资源简介:
DiQuID数据集是由亚里士多德大学电子与计算机工程学院和希腊信息技术研究所CERTH共同创建的高质量图像修复数据集。该数据集包含从MSCOCO、RAISE和OpenImages三个数据集中选取的78,684张原始图像生成的95,839张修复图像。数据集的创建采用了一种系统的方法,包括语义对齐的对象替换、多模型图像修复和不确定性指导的欺骗性评估三个主要组成部分,以确保图像修复的质量和多样性。该数据集旨在解决图像修复检测中缺乏大规模、高质量数据的问题,可用于训练和评估伪造检测算法。
The DiQuID dataset is a high-quality image inpainting dataset jointly created by the School of Electrical and Computer Engineering of Aristotle University of Thessaloniki and the Greek Information Technology Research Institute CERTH. This dataset includes 95,839 inpainted images generated from 78,684 original images selected from three existing datasets: MSCOCO, RAISE and OpenImages. The construction of the dataset adopts a systematic approach, which consists of three core components: semantically aligned object replacement, multi-model image inpainting, and uncertainty-guided deceptive evaluation, to ensure the quality and diversity of the inpainted images. This dataset aims to address the shortage of large-scale, high-quality data for image inpainting detection, and can be used for training and evaluating forgery detection algorithms.
DiQuID 数据集概述
数据集名称
DiQuID
数据集描述
DiQuID是一个大规模的由AI生成的图像修复基准数据集。
数据集用途
用于图像修复领域的评估和基准测试。
相关论文
"A Large-scale AI-generated Image Inpainting Benchmark"
代码仓库
代码可用性
代码将在不久的将来提供。




