Another milestone in Kakao Brain’s research on face swapping, the company’s paper “Smooth-Swap: A Simple Enhancement for Face-Swap with Smoothness” will be presented at the upcoming World Vision conference. By computer, CVPR 2022For the second year in a row. This will include an exclusive oral presentation session dedicated to the most outstanding papers among the accepted papers (25.33% of the 8,161 submissions accepted this year). At last year’s event, only 4% of accepted papers had time to give an oral presentation in which Kakao Brain was nominated for his outstanding research paper, “HOTR: Discovering End-to-End Human-Body Interaction with Transformers”. This year, “Smooth-Swap” has not only managed to significantly reduce the complexity of its architecture, but also has a significant market potential, which was recognized and awarded at the first conference on computer vision.
Accurate and consistent identity gradation Essential to change a person’s identity without sacrificing high image quality. “Smooth-Swap”, trained by supervised contrast loss, gained its stable identity by learning to merge more smoothly. These improvements address the weakness of the previous model of adding handcrafted components and 3D modeling of the face, which eventually complicated its design and led to advanced hyperparametric tuning. Instead, “Smooth-Swap” relies on a U-Net-based architecture with a built-in identity plugin to deliver industry-leading performance.
The simple structure and improved performance of “Smooth-Swap” not only made the technology competitive in terms of broader market potential and application, but also enabled it to handle face swap scenarios. More difficult, such as face swap during video playback. Smooth-Swap proposes a differentiated approach to identity integration and allows the creator to create high-quality images, especially when changing the subject’s face shape. Thanks to Kakao Brain’s “Smooth-Swap”, which enables fast and stable face exchange, the development of various types of digital humans such as virtual influencers, program hosts and broadcasters is expected.
“We are proud and excited to reveal our revolutionary face swap technology, Smooth-Swap, to the world,” said Kim Il-doo, CEO of Kakao Brain. “I firmly believe this technology will accelerate face-sharing innovation, bringing us closer to the incredibly immersive metaverse we have always dreamed of, and the digital human services of the future.”
About Cocoa Bren
Kakao Brain is a world-leading AI company with unparalleled AI technologies and R&D networks. The company was founded by Kakao in 2017 to solve some of the world’s biggest “unthinkable questions” through solutions made possible by life-changing AI technologies. Always at the forefront of innovation in the world of technology, Kakao Brain has developed many revolutionary AI services and models designed to improve the quality of life of thousands of people, including minDALL-E, KoGPT, CLIP/ALIGN and RQ-Transformer. As a global leader in artificial intelligence, Kakao Brain has the responsibility to foster a vibrant technology community and a robust R&D ecosystem as part of its mission to create new technology markets with unlimited potential. For more information, visit https://KakaoBrain.com/.
 Identity integration is a vector representation of a face image used to compare identities. If the representation vectors (or integration vectors) of two faces are close enough, their identities are considered identical.
 The CVPR (Computer Vision and Pattern Recognition Conference) has been co-sponsored by the Institute of Electrical and Electronics Engineers (IEEE) and the Computer Vision Foundation (CVF) since 1983, and is one of the most popular annual conferences in the computer vision sector, with the European Conference on Computer Vision (ECCV) and the International Computer Vision (ICCV).
 The identity gradient is a training signal that tells the face-switching model which part to adjust to precisely change a person’s identity.
Source: Kakao Brain