DeepMoireFake (DMF)
收藏资源简介:
DeepMoireFake (DMF) 数据集是一个包含 12,832 个视频的深度伪造数据集,这些视频来自 Celeb-DF, DFD, DFDC, UADFV 和 FF++ 数据集,涵盖了 35.64 小时的视频时长。数据集包含了在不同屏幕、智能手机、照明设置和相机角度下捕捉的视频片段,从而模拟了现实世界的各种条件。该数据集的创建旨在评估当前深度伪造检测器在面对 Moire 纹理时的鲁棒性,并推动未来研究朝着解决深度伪造检测中 Moire 引起的挑战的方向发展。
The DeepMoireFake (DMF) dataset is a deepfake dataset containing 12,832 videos sourced from Celeb-DF, DFD, DFDC, UADFV, and FF++ datasets, with a total runtime of 35.64 hours. It includes video clips captured under various settings including different screens, smartphones, lighting conditions, and camera angles, thus simulating diverse real-world conditions. This dataset is designed to evaluate the robustness of existing deepfake detectors against Moiré patterns, and to advance future research addressing the challenges posed by Moiré artifacts in deepfake detection.

- 1Through the Lens: Benchmarking Deepfake Detectors Against Moiré-Induced Distortions成均馆大学, 韩国CSIRO数据61, 澳大利亚 · 2025年



