What AI says about Localised-Deepfake-Detection

Official implementation of CVPR 2025 Localized Deepfake Detection, focusing on detecting localized deepfake manipulations in videos using action unit-guided video representations. The project includes setup instructions, required weights, test videos, and an inference script (inference.py) to output fake/real probabilities for videos. Users place model weights in the weights directory and run inference.py with a video path (e.g., assets/videos/) to obtain a probabilistic authenticity prediction.