DEEPFAKE CLI: Accelerated Deepfake Detection using FPGAs

• DOI: 10.48550/arxiv.2210.14743
• Publication Date: 2022-01-01
• Authors: Omkar Bhilare; Rahul Singh; Vedant Paranjape; Sravan Chittupalli; Shraddha Suratkar; Faruk Kazi

Because of the availability larger datasets and recent improvements in generative model, more realistic Deepfake videos are being produced each day. People consume around one billion hours video on social media platforms every day, thats why it is very important to stop spread fake as they can be damaging, dangerous, malicious. There has been a significant improvement field deepfake classification, but detection inference have remained difficult task. To solve this problem paper, we propose novel DEEPFAKE C-L-I (Classification-Localization-Inference) which explored idea accelerating Quantized Detection Models using FPGAs due their ability maximum parallelism energy efficiency compared generalized GPUs. In used light MesoNet with EFF-YNet structure accelerated VCK5000 FPGA, powered by state-of-the-art VC1902 Versal Architecture uses AI, DSP, Adaptable Engines for acceleration. We benchmarked our speed other nodes, got 316.8 FPS while maintaining 93\% Accuracy.