/Unmasking-Deepfake-Faces-from-Videos-An-Explainable-Cost-Sensitive-Deep-Learning-Approach-IEEE2023

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

Primary LanguageJupyter NotebookMIT LicenseMIT

Unmasking Deepfake Faces from Videos An Explainable Cost-Sensitive Deep Learning Approach

This repository contains the code and datasets used in the paper titled "Unmasking Deepfake Faces from Videos An Explainable Cost-Sensitive Deep Learning Approach" accepted and presented at the 26th International Conference on Computer and Information Technology (ICCIT) 2023.

Paper Link: PDF

Table of Contents

Dataset

We used publicly available datasets they are CelbDF-V2 and FaceForensics++

Result

Performance Metrics of Weighted Average on CelebDf-V2 Dataset

Model Accuracy Precision Recall F1 Score
XceptionNet 98% 0.98 0.98 0.98
InceptionResNetV2 0.97 0.97 0.97 0.97
EfficientNetV2S 0.97 0.97 0.97 0.97
EfficientNetV2M 0.97 0.97 0.97 0.97

Performance Metrics of Weighted Average on FaceForensics++ Dataset

Model Accuracy Precision Recall F1 Score
InceptionResNetV2 94% 0.94 0.94 0.94
XceptionNet 93% 0.93 0.93 0.93
EfficientNetV2S 92% 0.92 0.92 0.92
EfficientNetV2M 88% 0.89 0.88 0.88

Citation

If you found this code helpful please consider citing,

@inproceedings{mahmud2023unmasking,
  title={Unmasking Deepfake Faces from Videos Using An Explainable Cost-Sensitive Deep Learning Approach},
  author={Mahmud, Faysal and Abdullah, Yusha and Islam, Minhajul and Aziz, Tahsin},
  booktitle={2023 26th International Conference on Computer and Information Technology (ICCIT)},
  pages={1--6},
  year={2023},
  organization={IEEE}
}

License

This repository is licensed under the MIT License. See the LICENSE file for more information.