Spatio-Temporal Convolutional Neural Networks for Deepfake Detection: An Empirical Study
Vishal Kumar Sharma, Rakesh Kumar Garg, Quentin Caudron · 2023
As the creation of deepfakes becomes more prevalent and sophisticated, the need for accurate and robust detection methods intensifies. This paper presents a comprehensive empirical study on the efficacy of S patio-Temporal Convolutional Neural Networks (ST-CNNs) for deepfake detection. It explores how the rich spatio-temporal information contained within video frames can be exploited by ST-CNNs to distinguish between genuine and manipulated content. The study is underpinned by a robust testing framework, wherein a range of deepfake generation techniques are used to evaluate the detection model. It further investigates the effect of various layers and architectural elements on detection performance. The results demonstrate that ST-CNNs, by leveraging spatio-temporal correlations, can offer superior deepfake detection performance compared to the conventional CNN models. This work can guide the development of more efficient and effective deepfake detection strategies by providing empirical insights into the utilization of ST-CNNs.