PRISM_Benchmark
收藏资源简介:
PRISM(PhotoRealistic Image Synthesis and Manipulation)基准数据集是一个专门用于检测生成或合成图像的数据集,关联于研究论文《The PRISM benchmark: PhotoRealistic Image Synthesis and Manipulation to detect generated images》。该数据集主要用于图像分类任务,特别关注深度伪造检测、图像取证、AI生成图像和合成图像识别等领域。数据规模在1万到10万样本之间,其中真实图像部分来源于公开数据集COCO 2017(具体为验证集和训练集),需要用户自行下载。PRISM基准旨在为评估和推进照片级真实感图像合成与操纵的检测方法提供一个标准化的测试平台。
The PRISM(PhotoRealistic Image Synthesis and Manipulation)benchmark dataset is designed for detecting generated or synthetic images, associated with the research paper 《The PRISM benchmark: PhotoRealistic Image Synthesis and Manipulation to detect generated images》。It is primarily used for image classification tasks, with a focus on areas such as deepfake detection, image forensics, AI-generated image recognition, and synthetic image identification. The dataset size ranges from 10,000 to 100,000 samples. The real images in the dataset are sourced from the public COCO 2017 dataset (specifically the validation and training sets), which users need to download separately. The PRISM benchmark aims to provide a standardized testing platform for evaluating and advancing detection methods for photorealistic image synthesis and manipulation.
PRISM Benchmark 数据集概述
- 数据集名称:PRISM Benchmark
- 许可协议:CC-BY-4.0
- 任务类别:图像分类(image-classification)
- 标签:深度伪造检测、图像取证、AI生成图像、合成图像
- 样本规模:10K < n < 100K
数据集描述
该数据集源自论文《The PRISM benchmark: PhotoRealistic Image Synthesis and Manipulation to detect generated images》,专注于真实感图像合成与操作,用于检测生成图像。
数据集构成
- 真实图像:需从COCO数据集单独下载:
- 测试集:COCO 2017 Val(下载地址:http://images.cocodataset.org/zips/val2017.zip)
- 训练集:COCO 2017 Train(下载地址:http://images.cocodataset.org/zips/test2017.zip)
引用信息
bibtex @article{BARTOLUCCI2026104826, title = {The PRISM benchmark: PhotoRealistic Image Synthesis and Manipulation to detect generated images}, journal = {Computer Vision and Image Understanding}, pages = {104826}, year = {2026}, doi = {10.1016/j.cviu.2026.104826}, url = {https://www.sciencedirect.com/science/article/pii/S1077314226001931}, author = {Filippo Bartolucci and Samuele Salti and Giuseppe Lisanti} }




