AWARE-NET

PyTorch implementation of our paper "AWARE-NET: Adaptive Weighted Averaging for Robust Ensemble Network in Deepfake Detection". ICEPECC2025, IEEE, arxiv

Architecture Overview

1. Tier 1: Averages predictions within each architecture (Xception, Res2Net101, EfficientNet-B7) to reduce model variance.
2. Tier 2: Learns optimal weights for each architecture’s contribution through backpropagation, improving overall ensemble performance.

Configuration Options

• Dataset Fraction: Control the fraction of data used (default: 50%).
• Train/Val/Test Split: Default 70/15/15 split.
• Annotation Management: Options to force new splits or use cached annotations.

Training Pipeline

1. Train Individual Models: Start by training each model (Xception, Res2Net101, EfficientNet-B7) with/without augmentation.
2. Train Ensemble: Fine-tune the ensemble with pre-trained individual models.
3. Cross-Dataset Evaluation: Test model generalization across datasets.

Results

• FF++:

  • AUC: 99.22% (no aug.), 99.47% (aug.)
  • F1: 98.06% (no aug.), 98.43% (aug.)

• CelebDF-v2:

  • AUC: 100% (both)
  • F1: 99.94% (no aug.), 99.95% (aug.)

Cross-Dataset:

• AUC: 88.20% (FF++ → CelebDF-v2), 72.52% (CelebDF-v2 → FF++)
• F1: 93.16% (FF++ → CelebDF-v2), 80.62% (CelebDF-v2 → FF++)

Installation

1. Clone the repo:

   git clone https://github.com/recluzegeek/aware-net.git
   cd aware-net

2. Install dependencies:

   pip install -r requirements.txt

3. Download datasets (FF++ & CelebDF-v2), extract faces from videos and configure the respective paths in the config.py

Usage

To start the training and evaluation:

python main.py

Citation

@article{doi:10.1049/icp.2025.1162,
author = {Muhammad Salman  and Iqra Tariq  and Mishal Zulfiqar  and Muqadas Jalal  and Sami Aujla  and Sumbal Fatima },
title = {AWARE-NET: adaptive weighted averaging for robust ensemble network in deepfake detection},
journal = {IET Conference Proceedings},
volume = {2025},
issue = {3},
pages = {526-533},
year = {2025},
doi = {10.1049/icp.2025.1162},

URL = {https://digital-library.theiet.org/doi/abs/10.1049/icp.2025.1162},
eprint = {https://digital-library.theiet.org/doi/pdf/10.1049/icp.2025.1162},
abstract = { Deepfake detection has become increasingly important due to the rise of synthetic media, which poses significant risks to digital identity and cyber presence for security and trust. While multiple approaches have improved detection accuracy, challenges remain in achieving consistent performance across diverse datasets and manipulation types. In response, we propose a novel two-tier ensemble framework for deepfake detection based on deep learning that hierarchically combines multiple instances of three state-of-the-art architectures: Xception, Res2Net101, and EfficientNet-B7. Our framework employs a unique approach where each architecture is instantiated three times with different initializations to enhance model diversity, followed by a learnable weighting mechanism that dynamically combines their predictions.Unlike traditional fixed-weight ensembles, our first-tier averages predictions within each architecture family to reduce model variance, while the second tier learns optimal contribution weights through backpropagation, automatically adjusting each architecture's influence based on their detection reliability.Our experiments achieved state-of-the-art intra-dataset performance with AUC scores of 99.22\% (FF++) and 100.00\% (CelebDF-v2), and F1 scores of 98.06\% (FF++) and 99.94\% (CelebDF-v2) without augmentation. With augmentation, we achieve AUC scores of 99.47\% (FF++) and 100.00\% (CelebDF-v2), and F1 scores of 98.43\% (FF++) and 99.95\% (CelebDF-v2). The framework demonstrates robust cross-dataset generalization, achieving AUC scores of 88.20\% and 72.52\%, and F1 scores of 93.16\% and 80.62\% in cross-dataset evaluations. }
}