faceforensics

Towards deepfake detection that actually works

Code for Video Deepfake Detection model from "Combining EfficientNet and Vision Transformers for Video Deepfake Detection" presented at ICIAP 2021.

Determine whether a given video sequence has been manipulated or synthetically generated

Deepfakes Video classification via CNN, LSTM, C3D and triplets [IWBF'20]

Unofficial Implementation: Learning Self-Consistency for Deepfake Detection

A state-of-the-art, open-source deepfake detection system built with PyTorch and EfficientNet-B0, featuring a user-friendly web interface for real-time image and video analysis.

This repository contains our POC for a website which can easily check videos for manipulated areas. It was part of the Hackathon for Good in the Hague, 2019.

Deepfake video detection with eight EfficientNet-B4 checkpoints, Grad-CAM heatmaps and facial-region analysis.

Deepfake faces detection from forged videos where used explainable AI for models' robustness as well as cost sensitive methods for mitigating dataset imbalance problem

An Image Analyzer that can analyze and give Real or Fake predictions with confidence scores

This project was completed as part of the Deep Learning course (GLO-4030) at Laval University. Its goal is to detect deepfake videos using deep learning techniques on the FaceForensics dataset. We were able to achieve deepfake detection by using the EfficientNet model.

Deepfake Detection System using CV +DL

Repository for replicating the experiments described in the articles developed as part of the joint research between Universitas Mercatorum and Universidad de Cádiz (UCA).

Detect deepfakes by fusing spatial and spectral forensics in PyTorch — 0.90 video AUC on disjoint FF++/Celeb-DF/DFD splits. Weights, fixtures, and a live demo included.

Deepfake detection on FaceForensics++ using ResNet50, Xception(+SE), and Swin(+SE) with face crops vs full frames, frame sampling, aggregation, and Grad-CAM.

This repo is an enhanced toolkit with some updated methods processing original FaceForensics++ dataset.

Reproduction/refactoring of the FaceForensics++ classification process.

State-of-the-art forensic toolkit for deepfake detection. Features a full-network fine-tuned Xception model (98.5% accuracy) on FaceForensics++ and a Glassmorphic web UI.

Frequency-aware deepfake detection under lossy H.264 compression using XceptionNet, DCT features, learnable frequency masking, and cross-attention fusion.

Reproducibility artifacts for "A Controlled, Internally Audited Negative Result: Training Recipe, H.264 Compression Levels and a Frequency-Stream Deepfake Detector" (IEEE Access, under review): per-video scores, frozen decision rules, evaluation harness, claims-to-artifacts map. Archive (all versions): doi.org/10.5281/zenodo.22032528