The paper "Faster Than Lies: Real-time Deepfake Detection using Binary Neural Networks" by Romeo Lanzino, Federico Fontana, Anxhelo Diko, Marco Raoul Marini, and Luigi Cinque introduces a novel approach to deepfake detection using Binary Neural Networks (BNNs). The authors aim to address the challenge of real-time detection with minimal accuracy loss, which is crucial for combating the spread of deep-generated media that undermines trust in online content. Unlike previous methods that rely on large and complex models, their approach leverages BNNs, which offer significant efficiency gains in terms of computational resources and memory usage. The proposed method incorporates Fast Fourier Transform (FFT) and Local Binary Pattern (LBP) as additional channel features to enhance the detection of manipulation traces in the frequency and texture domains. Evaluations on three benchmark datasets—COCOFake, DFFD, and CIFAKE—demonstrate the method's state-of-the-art performance in most scenarios, achieving up to a 20× reduction in FLOPs during inference. The authors also conduct an ablation study to highlight the impact of each design choice, providing insights into the effectiveness of the proposed method. The contributions of this work include the first implementation of BNNs for deepfake detection, extensive experimentation on benchmark datasets, an ablation study, and quantitative results that promote further research in efficient deepfake detection. The paper concludes by discussing the limitations and future directions, emphasizing the potential for practical implementation on specialized hardware and the exploration of alternative pre-training datasets to enhance transfer-learning capabilities.The paper "Faster Than Lies: Real-time Deepfake Detection using Binary Neural Networks" by Romeo Lanzino, Federico Fontana, Anxhelo Diko, Marco Raoul Marini, and Luigi Cinque introduces a novel approach to deepfake detection using Binary Neural Networks (BNNs). The authors aim to address the challenge of real-time detection with minimal accuracy loss, which is crucial for combating the spread of deep-generated media that undermines trust in online content. Unlike previous methods that rely on large and complex models, their approach leverages BNNs, which offer significant efficiency gains in terms of computational resources and memory usage. The proposed method incorporates Fast Fourier Transform (FFT) and Local Binary Pattern (LBP) as additional channel features to enhance the detection of manipulation traces in the frequency and texture domains. Evaluations on three benchmark datasets—COCOFake, DFFD, and CIFAKE—demonstrate the method's state-of-the-art performance in most scenarios, achieving up to a 20× reduction in FLOPs during inference. The authors also conduct an ablation study to highlight the impact of each design choice, providing insights into the effectiveness of the proposed method. The contributions of this work include the first implementation of BNNs for deepfake detection, extensive experimentation on benchmark datasets, an ablation study, and quantitative results that promote further research in efficient deepfake detection. The paper concludes by discussing the limitations and future directions, emphasizing the potential for practical implementation on specialized hardware and the exploration of alternative pre-training datasets to enhance transfer-learning capabilities.