DEEPFAKE
DETECTOR
CYBERSECURITY
FORENSICS
AI/ML
THE_THREAT
With the proliferation of Generative Adversarial Networks (GANs), hyper-realistic deepfake videos have become a severe biometric vulnerability. Basic detection models fail against temporal warping.
ARCHITECTURE
I architected a Hybrid Neural Network. Extracting spatial features via ConvNets (Xception) and sequential temporal inconsistency traits via Recurrent LSTMs, creating an impenetrable detection pipeline.
PERFORMANCE
Achieved 94.8% validation accuracy on the deeply compressed Celeb-DF dataset, mitigating adversarial attacks by evaluating raw frame artifacts securely locally.
TECH_ARSENAL
PYTHON 3.10
TENSORFLOW / KERAS
OPENCV
PYTORCH
MATPLOTLIB