Compressed Deepfake Detection using Spatio-Temporal Approach with Model Pruning

Abstract

Deepfake is a deep learning technology that replaces source person face photos with the target person face photos in a movie to create a video of the target by executing the actions performed by source person. Due to the limited storage capacity and network bandwidth constraints, compressed media is now commonly employed in social networks. The goal of this research study is to determine whether or not a given compressed video is a deepfake. To do this, both the spatial and temporal components of the video should be considered, and the findings will be integrated by using an appropriate fusion approach. Hence, the deep learning model used in the spatial approach is pruned using the Network Pruning technique to achieve a better performance. The combined prediction of the spatial and temporal approaches indicates whether the given video is deepfake or not.