Kakao Brain brings major improvements to the metauniverse with new face-sharing technology

  • May 09, 2022 at 11:33 pm
  • Kakao Brain today announced the development of its newest face changing technology, Smooth-Swap, which is designed to expand the possibilities of seamless face changing through enhanced identity integration. The innovative model impresses with its innovative and simplified architecture after replacing complex external modules of the existing system with seamless integration of identities, which in turn enables faster and more stable face switching. Smooth-Swap: A Simple Enhancement to Face-Swapping with Smoothness marks another important milestone in Kakao Brain’s facial change research and will be presented for the second year in a row at the upcoming Global Computer Vision Conference, CVPR 2022. This includes exclusive oral sessions reserved for the most prominent articles among accepted articles (25.33% of 8,161 applications received this year). At last year’s event, just 4% of submissions received an oral presentation nominating Kakao Brain for its outstanding research, “HOTR: Discovering End-to-End Interaction with Transformers.” This year, Smooth-Swap not only significantly reduces its architectural complexity, but also has great commercialization potential, both of which have been recognized and appreciated by leading computer vision conferences. Precise and consistent identity gradients are essential for a smooth transition of one’s identity without compromising on high image quality. Trained through controlled loss of contrast, Smooth-Swap maintains its steady identity gradient by learning to integrate with greater fluency. This fix relates to the previous model’s weakness in adding handcrafted components and facial 3D modeling, which ultimately complicates the design and leads to complex hyperparameter adjustments. In contrast, Smooth-Swap relies on a simple U-Net based architecture with seamless embedded identity embedding to ensure the highest performance. The simplified Smooth-Swap architecture and improved performance not only make this technology competitive in terms of its commercialization potential and a wider range of applications, but also enable it to deal with more challenging face changing scenarios, such as: B. faces during video playback. Smooth-Swap offers a different approach to identity embedding and allows the generator to produce higher quality images, especially when changing the shape of a subject’s face. Kakao Brain’s “Smooth-Swap”, which enables fast and stable face swaps, aims to create different types of digital people, such as virtual influencers, presenters and broadcasters. “We are proud and excited to introduce this revolutionary smooth-swap technology to the world,” said Kim Il Sung, CEO of Kakao Brain. “I truly believe this technology will accelerate innovation in the face transformation field, bringing us one step closer to the highly immersive metaverse we’ve always dreamed of and the future of digital human services.