Deepfake Image Detection Using EfficientNetB4

This paper presented a hybrid EfficientNetB4 with Self-Attention framework for deepfake facial-image detection. The convolutional backbone captures hierarchical local evidence, while attention-based refinement models relationships among spatial regions. On the OpenForensics evaluation reported in the source thesis, the method achieved 84.33% test accuracy, 92.96% precision, 74.04% recall, 82.33% F1-score, 0.9253 ROC-AUC, and 0.9367 R-AUC. The results support the usefulness of commbining efficient CNN features with contextual attention, while the validation–test difference emphasizes that robust deployment requires broader, cross-domain evaluation.

[1] M. Tan and Q. V. Le, “EfficientNet: Rethinking model scaling for convolutional neural networks,” in Proc. Int. Conf. Mach. Learn. (ICML), 2019. [2] A. Vaswani et al., “Attention is all you need,” in Proc. Adv. Neural Inf. Process. Syst. (NeurIPS), 2017. [3] B. Dolhansky et al., “The DeepFake Detection Challenge (DFDC) dataset,” arXiv:2006.07397, 2020. [4] Y. Li, X. Yang, P. Sun, H. Qi, and S. Lyu, “Celeb-DF: A new dataset for deepfake forensics,” in Proc. CVPR, 2020. [5] A. Rossler et al., “FaceForensics++: Learning to detect manipulated facial images,” in Proc. ICCV, 2019. [6] Z. Yan et al., “DeepfakeBench: A comprehensive benchmark of deepfake detection,” in Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit. Workshops, 2023. [7] OpenForensics dataset and experimental configuration, as described in the source dissertation, “Deepfake Image Detection Using EfficientNetB4,” Jawaharlal Institute of Technology, Borawan, 2023–2026. [8] I. Goodfellow et al., “Generative adversarial nets,” in Proc. Adv. Neural Inf. Process. Syst., 2014. [9] K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,” in Proc. CVPR, 2016.