3D/4D scene representations
Efficient representations for geometry, appearance, motion, and time-varying real-world structure.
About
I am a Senior Scientist at ETH Zürich, working at the intersection of computer vision, computer graphics, and machine learning. My research focuses on representations of real-world 3D and 4D phenomena: reconstructing geometry, modelling motion, rendering photorealistic scenes, and extracting structure from visual observations.
I am increasingly interested in how these representations can support spatial intelligence: the ability of models to form, use, and update internal maps of the world for reasoning, planning, interaction, and understanding.
Research
Neural reconstruction and rendering
Methods for reconstructing and rendering photorealistic static and dynamic scenes from sparse or imperfect observations.
GGPT · MVTracker · SplatFormer · SplatFields
Digital twins
Digital representations of humans, biological systems, and surgical scenes: controllable, temporally coherent, and grounded in real observations.
SurgiScope · Surgical Digital Twins · Morphable Diffusion · HARP · SMPLpix
Spatial intelligence
Understanding what kinds of internal maps vision models need for reasoning about space, objects, motion, navigation, interaction, and change.
Current direction
I see spatial intelligence as a bridge between geometric reconstruction and higher-level reasoning. A useful visual system should not only recover shape, motion, and appearance; it should maintain internal maps that support questions such as what is where, what changes, what can move, and what enables planning or interaction.
Teaching and supervision
At ETH Zürich, I co-teach Computer Vision (with Marc Pollefeys and Siyu Tang) and Digital Humans (with Siyu Tang).
I supervise student projects in computer vision, 3D reconstruction, neural rendering, and spatial intelligence, with particular interest in work that connects rigorous geometric modelling with modern foundation models.
If you are a student interested in working on these topics, please reach out. The most useful messages are short and specific: what you want to work on, why it interests you, and what you have already tried.