The authors of Faceless, the Make-in-Vietnam Deepfake detection toolkit, are two white-hat hackers, Duong Tieu Dong, born in 2005, and Pham Tien Manh, born in 1996.

According to the development team, cybercriminals can use Deepfake technology to commit fraud. There have been many studies about tools to detect Deepfake, but they remain unpopular or limited, and only used for commercial purposes. 

That is why the team decided to conduct research and develop a toolkit to improve  awareness and help users identify fraudulent behavior and the use of Deepfake.

Pham Tien Manh, the white hat hacker, said Faceless operates by receiving an image from users and performing facial recognition. The face is put into a convolutional neural network (CNN) for classification to find out if it is a product of a Deepfake.

“At first, we built some CNNs based on existing models by using transfer learning. After that, they were compared with each other to find the most optimal model. From this, the neural network is used to build the system to detect Deepfake images,” Duong Tieu Dong explained.

It took the team half a year to complete the products, from theoretical study to application in practice. The toolkit is shared under an open source code.

Experiments show that Faceless can discover Deepfakes after less than two seconds with the accuracy rate of 94.5 percent. Other toolkits available internationally have a higher accuracy rate of 97 percent in discovering Deepfake.

According to Manh, thanks to the outstanding results in detecting Deepfake compared with existing tools, the development team was invited to BlackHat Asia 2023 forum to give a presentation about the tool.

At BlackHat Asia 2023, the organizing board created favorable conditions for the team of Manh and Dong to use the laboratory area with 10 laptops to allow visitors to experiment the tool and discover Deepfakes.

With more than one hour of presentation, the team shared experiences on how to use deep learning to detect Deepfakes, and then conducted a demo with Faceless.

“After the presentation, we realized that the organizing board and attendees had a good impression about the toolkit. They also showed interest by suggesting improvements to make the toolkit more effective,” Manh said.

Manh said there are still many things that need to be done, because the toolkit is just the result of laboratory research and has not been used in reality.

Trong Dat