Byte-identical copies of published model files. They exist only because the SkillSafe browser-model vetting intake accepts Hugging Face, npm and GitHub URLs (pinned to a commit) and not the publisher's own storage host, and vets single ONNX / WebAssembly / TFLite files rather than zip bundles. Verify any file by hashing it: the sha256 must match this table and the original.
| File | Extracted from | sha256 | Bytes |
|---|---|---|---|
mediapipe/face_landmarker/float16/1/face_detector.tflite |
https://storage.googleapis.com/mediapipe-models/face_landmarker/face_landmarker/float16/1/face_landmarker.task, zip entry face_detector.tflite |
b4578f35940bf5a1a655214a1cce5cab13eba73c1297cd78e1a04c2380b0152f |
229,746 |
mediapipe/face_landmarker/float16/1/face_landmarks_detector.tflite |
same bundle, zip entry face_landmarks_detector.tflite |
c7d54204ce0448474c7f3fa9af494787c0965cbdd6f20fc72867e43046bd43d5 |
2,553,590 |
The bundle itself is sha256 64184e229b263107bc2b804c6625db1341ff2bb731874b0bcc2fe6544e0bc9ff
(3,758,596 bytes); unzip it and hash the two entries to reproduce the table.
Licence: Apache-2.0 (Google MediaPipe, https://github.com/google-ai-edge/mediapipe/blob/master/LICENSE).
Model cards: BlazeFace short-range and Face Mesh V2, at storage.googleapis.com/mediapipe-assets.
Used by https://id-photo-maker.skillsafe.ai/ for its on-device head-height and eye-line check,
which packs the two files back into a stored zip in the browser and runs them with the
MediaPipe Tasks Vision runtime (@mediapipe/tasks-vision 1.0.1).