This project considers immersive and structurally accurate modelling in computer vision under real-world constraints on consumer-grade hardware. We fuse camera motion estimation with geometric reconstruction and immersive deep modelling in a robust end-to-end fashion. The project leverages probabilistic methods and develops both data- and compute-efficient approaches. These are of particular importance in both non-human perception that requires accurate structural knowledge and in immersive technologies such as augmented and virtual reality (AR/VR) that require both structural accuracy and photorealism.

2024

2028