Show & tell: fully client-side 2D→SBS 3D video converter — our own compact depth model in the browser (ONNX, ~8MB)
Hi all — long-time lurker, first show & tell here.
We built depthsnap.com — a 2D→SBS (side-by-side 3D) video converter that runs entirely in the browser: files never leave the device, no upload, no account.
Under the hood it runs our own compact monocular-depth model exported to ONNX (~8MB, ~50ms/frame on a mid-range phone GPU) via ONNX Runtime Web — distillation-first training, quantization-friendly, built for exactly this "small model, big audience" use case the community keeps discussing.
Also on the site:
- A plain-English full-SBS vs half-SBS explainer (the most common confusion we see)
- Per-device playback guides: Quest / Vision Pro / Pico / Xreal
- Troubleshooting pages for depth artifacts & playback judder
It's early and the depth-quality defaults are our best guess so far. If you try it on real footage, feedback on depth strength/artifacts would be genuinely appreciated — especially on fast motion and low-texture scenes, which are our known weak spots.
— an indie dev (@greatYiTin)