基于多任务学习和注意力机制的双分支深度换脸检测方法

深度换脸技术的出现严重威胁了公众的隐私安全。为了解决现有深度换脸检测方法的局限性，基于多任务学习策略提出了一种双分支检测网络，实现在检测视频伪造的同时逐帧检测。该网络引入了注意力机制和时序学习模块，通过学习局部空间信息和时序信息提升检测性能。该方法在公开数据集Celeb-DF和FaceForensics++上获得了比当前先进换脸检测方法更高的准确率和ROC(Receiver Operating Characteristic)曲线下面积（Area under ROC Curve,AUC），面对不同光照、人脸朝向、视频质量时表现出了良好的鲁棒性。

Two-branch deepfake detection method based on attention mechanism and multitask learning

performing frame-by-frame detection.The attention mechanism and temporal learning module are introduced to learn local spatial and temporal information to effectively improve performance.Experiments on public datasets Celeb-DF and FaceForensics++ show that the proposed method achieves higher accuracy and the Area under ROC Curve(AUC) scores than state-of-art methods.Furthermore,this method shows better generalization robustness on different light conditions,face orientations,video qualities than existing methods.