Researching Deepfake Detection Alternatives

ABSTRACT:

In today's world, accurate Deepfake detection is in high demand as we witness a significant rise in incidents with ever-increasing severity and reach spreading in a variety of media. To combat this, complex detection models are being developed constantly but require extensive amounts of computational power in order to function properly. I attempted to create a computationally light model by using Google Colab and a Risk-Engine-based model. Using different mathematical functions that are built into various libraries in Python, I was able to create a functioning model that utilizes different modalities such as texture analysis. The model was able to give an average training accuracy score of 65% and an average testing accuracy score of 60%. In order to further this model, we could try adding more modalities or by developing a different new model. 

Writing outline.docx

Download DOCX • 22KB