Head over to our on-demand library to view sessions from VB Transform 2023. Register Here
Digital identity and 3D graphics company Itseez3D today announced the launch of Avatar SDK, a deep fake detector platform that aims to enable businesses to fortify user security and bolster application integrity.
In response to the increasing prevalence of synthetic avatars and deepfake technology, verifying user identities has become paramount for developing secure and trustworthy applications. Itseez3D said that it recognized this urgency and developed the Avatar SDK Deep Fake Detector to uphold the authenticity of user identities.
The platform empowers facial verification systems and digital identity management platforms with tools to combat the escalating number of fraudulent attempts. By leveraging machine learning (ML) algorithms, the platform can analyze facial features, distinguishing between genuine photos and synthetic 3D avatars.
Trained on real photos and avatar renderings
The company claims that traditional deepfake detectors focused solely on detecting images created through neural rendering, or when neural networks directly synthesize images. However, these detectors were insufficient in identifying 3D models rendered with a traditional 3D graphics toolchain.
VB Transform 2023 On-Demand
Did you miss a session from VB Transform 2023? Register to access the on-demand library for all of our featured sessions.
“Our machine learning approach trains on real photos and avatar renderings to detect deepfakes as opposed to traditional facial recognition systems that mostly look at the inner part of the face (the area surrounding the eyes, nose, and mouth),” Victor Erukhimov, founder and CEO of Itseez3D, told VentureBeat. “We analyze the whole head image, including hair and neck.”
Avatar SDK aims to ensure that only legitimate users gain access to a given platform by identifying inconsistencies and markers characteristic of deepfakes.
Compatible with diverse platforms
Erukhimov stated that the company has packaged Avatar SDK as a Docker container, facilitating its integration into enterprise applications and deployment on organizational servers. Moreover, the SDK is compatible with diverse platforms, including social networking apps, e-commerce platforms and immersive gaming platforms.
“We have deployed our tech as a Docker container, as it enables our customers to inculcate the solution in the same cloud where data is processed, addressing the privacy concerns,” Erukhimov told VentureBeat. “The Deep Fake Detector analyzes all data in the customer’s cloud, ensuring data doesn’t leave the customer’s storage environment.”
Tackling the rise of deepfakes through machine learning
Erukhimov revealed that an incident in early January 2023 triggered the product’s inception. The company noticed an unusual surge in traffic to their avatar creation demo from Bangladesh.
Malicious operators were exploiting the demo via YouTube videos to bypass Bangladesh’s facial verification system for the National Identity Card (NID). Although the avatars were not hyper-realistic, they managed to deceive detection systems, raising concerns about potential voter fraud during the upcoming presidential elections.
In response, Itseez3D took proactive measures by blocking Bangladesh’s IPs, notifying the government, and offering a free avatar deepfake detector. Recognizing the significance of this solution for other organizations, Itseez3D subsequently developed the Avatar SDK.
“We believe that digital identity is very important in today’s world where so much is based on digital verification, including voting and online payments,” said Erukhimov. “Reconstruction of avatars from multiple images or videos helps create a geometrically accurate avatar that malicious actors use today to access bank accounts. Our Deep Fake Detector provides a detection accuracy of over 99% with a false alarm rate under 2% to mitigate such issues and protect confidential data.”
What’s next for Itseez3D?
Itseez3D stated that it is presently working on creating human-like game-ready avatars from selfies. Additionally, it has formed partnerships with VR developers including Reallusion and Spatial and VR games like Drunkn Bar Fight, to incorporate these avatars into their products.
“We are now working on the next generation of avatars created from selfies that we call MetaPerson,” Erukhimov told VentureBeat. “It takes under a minute to create an avatar that looks like you through a selfie. We are already working with a few existing customers on the integration and believe that these avatars will enable many use cases, including being yourself in an AR/VR game/experience, Metaverse, e-commerce and more.”