Post by National AI Research Lab (NAIRL)
1,213 followers
"12 papers at #CVPR and 5 papers at #ICLR in a single year. A historic milestone for a single research lab in Korea!" We are thrilled to share that Prof. Seungryong Kim's team (#CVLAB) at KAIST AI, a key partner of the National AI Research Lab (NAIRL), has once again proven its world-class capabilities in computer vision and generative AI. Through massive joint research with global top-tier institutions and tech giants like SonyAI, Adobe Research, University of California, Berkeley, ETH Zürich, NAVER Corp AI Lab, and Samsung Electronics, the team has introduced groundbreaking technologies ready to transform everyday life and industries. 💡 Key Research Highlights: 1️⃣ Perfect Control in Next-Gen Generative AI • APPLE (with Samsung Electronics): A novel diffusion framework for face swapping that perfectly preserves fine target-specific attributes often lost in existing models, such as pose, expression, lighting, and makeup. • 3D Scene Prompting (with Sony AI): Enables future video generation with freely controllable camera viewpoints while maintaining strict 3D scene consistency and temporal stability. 2️⃣ Hyper-Efficiency in 3D Vision • C3G (with Sony AI, ETH Zurich): Instead of using massive data for 3D reconstruction, this work achieves high-quality 3D rendering and scene understanding using a highly compact set of only around 2,000 Gaussians. 3️⃣ AI for the Upcoming Robotics Era • ORCA (with Naver AI Lab): Integrates diffusion models to generate task-adaptive visual representations for robotic control, enabling robots to flexibly recognize and respond to situations based on visual information. A massive congratulations to Prof. Seungryong Kim and the brilliant researchers at CVLAB for elevating Korea's AI technology to the global forefront and laying a solid technical foundation for the era of embodied AI and robotics! 👏 🔗 Explore the full list of publications and projects: https://cvlab.kaist.ac.kr/ #NAIRL #KAIST #CVLAB #SeungryongKim #CVPR2026 #ICLR2026 #ComputerVision