Show HN: Apple's Sharp Running in the Browser via ONNX Runtime Web

· web ai open-source · Source ↗

TLDR

  • Browser-based Gaussian splat generator running Apple’s SHARP model via ONNX Runtime Web, outputting downloadable .ply files.

Key Takeaways

  • Single-image input generates Gaussian splats entirely client-side; no server inference required.
  • ONNX export produces two files (sharp_web_predictor.onnx + .onnx.data); both must be co-served, total ~2.4 GB.
  • Stack: React, TypeScript, Vite, Bun, ONNX Runtime Web WASM, GaussianSplats3D viewer; Chrome/Edge desktop recommended.
  • Apple’s SHARP model weights carry a research-use-only license (LICENSE_MODEL) separate from the code license.
  • Performance is heavy: first run is slow due to model and WASM runtime init; low RAM or mobile will likely crash.

Hacker News Comment Review

  • No substantive HN discussion yet.

Original | Discuss on HN

Related coverage