Computer vision kernels racing side by side. A webcam feed or a rendered 3D scene flows through WebGPU compute passes: reference implementations against optimized variants, with GPU timestamp latency and a pixel-identical correctness badge.
What it will demonstrate
Workgroup-tiled gaussian blur vs the naive per-pixel pass, with the memory-access story visualized
A full multi-pass Canny chain: blur, gradient, non-max suppression, hysteresis
Harris corners computed on the live feed and overlaid in real time
The base-vs-optimized latency delta, measured per pass
Roadmap
Milestone 1: luminance and gaussian (base vs workgroup-tiled)
Milestone 2: Sobel and the Canny chain
Milestone 3: Harris corners on the live feed
Milestone 4: research-grade optical flow pass
Questions or collaboration: contact.
Progress notes land on the main page as milestones ship.