FACEtMASK

360° equirect video → reframed pinhole views, optionally SAM 3 masked.

internal build

Standalone tool that turns a 360° capture into a folder of pinhole-projection images (with optional per-view masks) ready for Metashape, LichtFeld Studio, Postshot, or any structure-from-motion / 3DGS pipeline. Ordinary photos and non-360 video are auto-detected and passed straight through.

Download (.zip) View on GitHub

Install (Windows)

  1. Download and unzip, or clone: 
    git clone https://github.com/flytbox-actual/facetmask.git
  2. Double-click install.bat (sets up Python 3.12 + a private virtual env, installs PyTorch + facetmask, verifies the GPU).
  3. Launch:
    • launch-app.bat — native desktop app (recommended)
    • launch-gui.bat — browser UI (cross-device)
GPU PyTorch is auto-selected. install.bat detects your GPU and installs the matching CUDA build — cu128 for RTX 50-series (Blackwell), cu124 for older NVIDIA cards, CPU otherwise. If you ever installed PyTorch by hand and SAM 3 errors with “no kernel image is available” on a 50-series card, fix it with: 
pip install torch torchvision --index-url https://download.pytorch.org/whl/cu128 --no-cache-dir

Updating

The app checks this site for new versions on launch and prompts you when one is available. To update manually, run update.bat (or re-run install.bat — it upgrades in place).

Requirements

Windows 10/11 · NVIDIA GPU with CUDA for SAM 3 masking (runs on CPU without, slowly) · a Hugging Face token with access to facebook/sam3 for masking · ~5 GB disk for PyTorch.

Links

Changelog Repository Docs (README) Roadmap