Celeb-DF
深度伪造检测领域的重要基准数据集，2020年由李云鹏等人发布，包含超过5000段高质量换脸视频及对应真实视频，伪造质量显著高于早期FaceForensics++数据集，专为评估深度伪造检测模型设计
Timeline (last 90 days)
The study achieved an average accuracy of 97.79% by processing 15 videos in parallel on the Celeb-DF dataset.
The original authors emphasized that this accuracy reflects the result of a specific system on a specific benchmark, and is not a claim of general robustness.
Celeb-DF was released in 2020 by Li Yunpeng et al., containing over 5,000 face-swap videos with refined post-processing and their corresponding real original videos.
The forgery quality in the Celeb-DF dataset is significantly higher than that in the earlier FaceForensics++ dataset, and it was designed to address the problem of early benchmarks being too easy.
All Facts (4)
The study achieved an average accuracy of 97.79% by processing 15 videos in parallel on the Celeb-DF dataset.
50%UnverifiedThe original authors emphasized that this accuracy reflects the result of a specific system on a specific benchmark, and is not a claim of general robustness.
50%UnverifiedCeleb-DF was released in 2020 by Li Yunpeng et al., containing over 5,000 face-swap videos with refined post-processing and their corresponding real original videos.
50%UnverifiedThe forgery quality in the Celeb-DF dataset is significantly higher than that in the earlier FaceForensics++ dataset, and it was designed to address the problem of early benchmarks being too easy.
50%