WildFake
收藏资源简介:
WildFake数据集是由蚂蚁集团和上海交通大学清远研究院共同创建的,旨在评估AI生成图像检测的泛化性和鲁棒性。该数据集收集了来自开放源社区的大量假图像,涵盖了广泛的图像类别和风格。WildFake包含由不同类型的生成器(如GANs、扩散模型等)生成的图像,其层次结构设计有助于提升检测器的性能。此外,数据集还包含了真实图像,以确保检测器在多种实际场景中的有效性。WildFake的应用领域包括隐私保护、真实性验证和安全性增强,旨在解决AI生成图像带来的潜在风险。
The WildFake Dataset was jointly created by Ant Group and Qingyuan Research Institute of Shanghai Jiao Tong University, aiming to evaluate the generalization and robustness of AI-generated image detection models. This dataset collects a large number of fake images from open-source communities, covering a wide range of image categories and styles. WildFake contains images generated by various types of generators such as GANs, diffusion models, etc., and its hierarchical structure design helps improve the performance of detectors. In addition, the dataset also includes real images to ensure the effectiveness of detectors in diverse practical scenarios. The application fields of WildFake cover privacy protection, authenticity verification and security enhancement, targeting the resolution of potential risks brought by AI-generated images.

- 1WildFake: A Large-scale Challenging Dataset for AI-Generated Images Detection蚂蚁集团 · 2024年



