Deepfake Detection System
TrueSight is a deep-learning–powered system for detecting deepfake videos. Built using PyTorch, the project combines powerful ResNeXt50 visual features with an LSTM sequence model to identify manipulated video content with high accuracy. It includes the complete pipeline — preprocessing, training, evaluation, and visualization.
● Designed a multi-stage pipeline using ESRGAN, MTCNN, ResNeXt50, and LSTM for high-accuracy video analysis. ● Integrated super-resolution, face alignment, spatial feature extraction, and temporal modeling for robust detection. ● Achieved 0.926 AUC, 85% recall, and 83% F1-score on the validation set across benchmark datasets.