Dynamic Video

Dynamic video research focuses on developing methods to analyze, manipulate, and synthesize videos containing moving objects and changing scenes. Current efforts concentrate on improving video restoration techniques (e.g., addressing atmospheric turbulence), generating realistic video content from various inputs (still images, sparse annotations, or even text), and accurately estimating 3D scene dynamics and object poses from video sequences. These advancements leverage neural networks, including diffusion models, transformers, and radiance fields, to achieve high-fidelity results and enable applications ranging from enhanced video editing and special effects to improved medical image analysis and robotics.