Unmasking Digital Deception
This project aims to identify deepfake technology in videos by using deep learning techniques. Through this program, users can validate the authenticity of videos in less than 10 seconds irrespective of the total duration of the video. Stay safe and keep aware!
Leveraging deep learning techniques, our tool meticulously analyzes intricate patterns within video frames using the Resnet50 architecture, and follows it up by a thorough anomaly detection process by the use of autoencoder reconstructions. Unlike traditional anomaly detection systems that look for high errors (indicating something is 'out of place'), this model has uniquely learned patterns where deepfakes often exhibit surprisingly low reconstruction errors. This model has been trained on a grand total of 180 videos and has potential for further improvement in the future.
Simply upload your video (MP4, AVI, MOV, MKV) and let the AI do the heavy lifting. In a few seconds, the AI will provide an instant analysis, helping you discern real from fake.