This project explores how computers can better understand and model the three-dimensional world using images and videos. We aim to develop intelligent systems that combine geometric structure, visual meaning, and immersive appearance into a unified 3D model. The research focuses on creating accurate and efficient methods that work without laborious manual calibration, making them easier to apply in practice. To achieve this, the project leverages large-scale learning powered by high-performance computing. These technologies are especially relevant for applications like autonomous vehicles, robotics, and augmented reality, where systems must understand complex real-world environments reliably and possibly in real time.

2026

2028