Sources & licenses
Face Shape Detector credits & licenses
The Face Shape Detector was built for LoschaZ. The face landmarks are found in your browser with open-source software and an open model from Google’s MediaPipe project:
What we built
The workspace, adding photos by drag and drop, file picker, selfie camera and clipboard, the roll correction, the hairline estimate and its adjustable handle, the draggable measuring points with the magnifier, the glasses try-on that draws frame shapes on the photo, the measurements and ratios, the classifier with its prototype table, the photo checks (pose, size, light, sharpness), the iris-based size estimate and frame width, the manual measuring form, the recommendations for hairstyles, glasses and beards with the frame drawings, the result image, and the packaging of the engine on this website were written for LoschaZ.
Components & licenses
- MediaPipe Tasks Vision (@mediapipe/tasks-vision 1.1.1) by Google, with its WebAssembly builds (with and without SIMD) – Apache License 2.0 (license and notice). Served from this website, with the change described below.
- MediaPipe Face Landmarker model bundle (face_landmarker.task, float16, version 1): the BlazeFace short-range face detector, Face Mesh V2 with 478 landmarks and the blendshape model – Apache License 2.0 according to its model card (model overview and model cards). The model card notes that the landmarks do not provide facial recognition or identification. Served from this website, unchanged.
Our changes
MediaPipe Tasks Vision contains a usage logger that would send usage statistics to Google every minute. In our copy of vision_bundle.mjs this logger is switched off, so nothing is queued or sent and the tool only loads its own files; the source map reference is removed as well. The file starts with a comment that states these changes (Apache License 2.0, section 4(b)). The WebAssembly files and the model are used unchanged.