Scalable Approaches for Visual Scene Understanding

National AI Research Lab (NAIRL) hosted an online invited talk by Dr. Junhwa Hur, Research Scientist at Google DeepMind, on April 11, 2025.

Under the theme “Scalable Approaches for Visual Scene Understanding,” Dr. Hur shared perspectives on how visual scene understanding, spanning tasks such as motion and geometry estimation, can be advanced through approaches that scale with data and models rather than relying on task-specific designs. The talk introduced directions in which large-scale learning can push the accuracy and generality of core computer vision capabilities.

Dr. Hur is a computer vision researcher at Google DeepMind whose work spans optical flow, scene understanding, and generative approaches to vision, including research demonstrating that diffusion models can achieve state-of-the-art performance in optical flow and monocular depth estimation. He received his Ph.D. from Technische Universitat Darmstadt under Professor Stefan Roth, his M.Sc. from Seoul National University, and his B.Sc. from POSTECH. NAIRL will continue to connect Korea’s research community with leading AI experts from around the world through its invited talk series.