Kenneth Xu

Research Interests

My ultimate vision is a world enabled by AI where anyone can travel and perceive anywhere in the world without being physically present—imagine visiting the Roman Colosseum and witnessing gladiatorial combat unfold before your eyes, or exploring the depths of the Amazon rainforest with perfect fidelity. This dream drives my research in 3D scene understanding and semantic representation, where I develop foundational technologies that make such immersive experiences possible.

Through my work on MastSAM and Splat Feature Solver, I'm building the technical foundations for this vision. MastSAM solves the critical challenge of multi-view inconsistency in 3D segmentation by leveraging 3D coordinate mapping techniques, ensuring coherent spatial understanding across different perspectives—essential for creating seamless virtual environments. Meanwhile, Splat Feature Solver enables rich semantic understanding of 3D scenes by efficiently lifting powerful image descriptors like CLIP into 3D Gaussian Splat representations, allowing AI systems to understand and interact with virtual worlds at a semantic level.

These technologies converge toward creating photorealistic, semantically-aware virtual worlds where users can not only see but truly understand and interact with their surroundings. My research in Human-Computer Interaction complements this vision by designing intuitive interfaces that make these complex 3D environments accessible to everyone, bridging the gap between cutting-edge AI capabilities and natural human interaction.

I'm always eager to connect with fellow researchers, industry professionals, or anyone passionate about making virtual reality indistinguishable from physical reality. Whether you're interested in collaborating on 3D scene understanding, semantic representation, or building the interfaces that will define our virtual future, I'd love to hear from you!