Neuroglancer WebGPU Prototype

This is an independent Vite React prototype for a WebGPU-backed, x/y-plane-only precomputed image volume viewer.

Run

npm install
npm run dev-server

Use a browser with WebGPU and OffscreenCanvas support.

Scope

• One precomputed image layer only.
• Public gs://, http://, and https:// precomputed roots.
• Frontend-side source normalization and metadata validation inside the ViewerEngine API boundary.
• Worker-side chunk selection, chunk download/decode scaffolding, resource scheduling, cache accounting, and WebGPU rendering.
• Drag pans x/y.
• Wheel changes z.
• Ctrl + wheel performs cursor-centered zoom.
• URL state stores only source, center, and zoom.

Architecture

React is the first UI, but not the visualization boundary. Frontends interact with src/client/ViewerEngine.ts. The default WorkerViewerEngine validates metadata before sending normalized source metadata to the worker. React components must not import worker, precomputed worker-side, decoder, cache, scheduler, or WebGPU modules directly.

The prototype intentionally does not depend on the local Neuroglancer package. Required precomputed behavior is re-implemented locally from the format behavior.

Verify

npm run typecheck
npm run test
npm run lint:boundaries
npm run build