GLFE: Video Frame Interpolation Forensics via Multi-Scale Global and Local Feature Exploitation

With the continuous development of video technology, video frame interpolation (VFI) has made significant advances. Frame interpolation can generate new frames from existing ones to increase the frame rate. At the same time, it poses a serious threat to information security. As the videos produced by frame generation technology become increasingly realistic, many of them can mislead the public. Therefore, an algorithm that is robust and highly accurate is needed to determine whether a video is authentic. In this paper, we construct a video frame interpolation forensic framework. On one hand, we propose a multi-scale framework that processes video features from coarse to fine. On the other hand, we combine a global information module and a local information module to enrich the extracted tampering features. Additionally, we place an efficient multi-scale attention (EMA) module at the end of our framework to fuse the results from the three scales. Extensive experiments demonstrate that our method achieves comparable performance to the current state-of-the-art detectors.