ForensicHub
收藏资源简介:
ForensicHub是一个统一的基准和代码库,旨在解决所有领域内的假图像检测和定位问题。该数据集涵盖了深度伪造检测、图像操作检测和定位、人工智能生成图像检测和文档图像操作定位四个领域,包含了23个数据集、42个基线模型、6个主干网络以及11个图像和像素级评估指标。ForensicHub通过模块化和配置驱动的架构,实现了跨数据集、转换、模型和评估器的可互换组件,允许灵活地组合所有领域的模型和数据进行训练和测试,为研究人员提供了深入的见解,并促进了未来在假图像检测和定位领域的突破。
ForensicHub is a unified benchmark and codebase designed to address the problem of fake image detection and localization across all domains. It covers four research areas: deepfake detection, image manipulation detection and localization, AI-generated image detection, and document image manipulation localization. The benchmark includes 23 datasets, 42 baseline models, 6 backbone networks, and 11 image-level and pixel-level evaluation metrics. Adopting a modular and configuration-driven architecture, ForensicHub provides interchangeable components across datasets, data transformations, models and evaluators, enabling flexible combination of models and data from all domains for training and testing. It offers researchers in-depth insights and facilitates future breakthroughs in the field of fake image detection and localization.




