Adversarial Dataset for Image Forgery Localization

We present an adversarial dataset tailored for evaluating the robustness of image forgery localization models. This dataset includes adversarially crafted forged images and their corresponding pixel-level forgery masks. Adversarial images are generated by attacking four state-of-the-art forgery localization models \u2014 including MVSS-Net, IF-OSN, HDF-Net and CoDE \u2014 across four widely-used datasets: Casia v1, IMD20, Columbia, and MISD.The adversarial images are designed to preserve perceptual quality while significantly degrading forgery localization performance.  This dataset enables systematic benchmarking of forgery localization robustness under white-box attacks, supporting both academic research and practical evaluation of secure forensic models.