Perception-aware 3D Gaussian Splatting

VR and AR headsets have to render rich 3D scenes fast enough to feel real, but the eye only resolves fine detail at the center of gaze. This project explores rendering and reconstruction that follow human perception, spending detail where it matters and saving it everywhere else.

Reconstruction and rendering guided by how we see

I’m working on this as a Summer Directed Study at NYU Tandon’s Immersive Computing Lab under Prof. Qi Sun, focused on perception-aware 3D Gaussian Splatting and foveated rendering for immersive displays.

What if a scene only rendered detail exactly where the eye is looking?

Where the work is heading

The goal is pipelines that stay sharp where you look while cutting the cost of the periphery, so immersive scenes can be both faithful and fast on headset-class hardware. This is ongoing research.