README.md

---
license: apache-2.0
tags: [onnx, mediapipe, blendshapes, arkit, face-capture, qtmesheditor]
---

# Face blendshapes MLP-Mixer (ONNX)

Google MediaPipe's face blendshape model (MLP-Mixer), converted to ONNX.

- **Input** `[1,146,2]` — the 146-landmark subset as **pixel coords**
  (`x·imgW, y·imgH`).
- **Output** `[52]` — **52 ARKit-compatible blendshape scores** in `[0,1]`,
  in MediaPipe order (`_neutral`, `browDownLeft`, … `jawOpen`, …
  `mouthSmileLeft/Right`, … `eyeBlinkLeft/Right`, …). These drive ARKit-style
  morph targets directly.

Converted with tf2onnx optimizers disabled (the optimized graph broke a
LayerNorm under ONNX Runtime — see the spike doc).

## License

Apache-2.0 — this graph is a direct ONNX conversion of a Google
[MediaPipe](https://developers.google.com/mediapipe) model (Apache-2.0 code
AND weights). Conversion + numerical-parity proof (vs the Python `mediapipe`
reference): [`scripts/export-facecap-onnx.py`](https://github.com/fernandotonon/QtMeshEditor/blob/master/scripts/export-facecap-onnx.py),
contract in [`docs/MOCAP_SPIKE.md`](https://github.com/fernandotonon/QtMeshEditor/blob/master/docs/MOCAP_SPIKE.md).

## How it is used

Mirror of one graph from [`fernandotonon/QtMeshEditor-models`](https://huggingface.co/fernandotonon/QtMeshEditor-models)
(`mocap/…`), which [QtMeshEditor](https://github.com/fernandotonon/QtMeshEditor)
downloads on first use for its **Performance Capture** feature (video/webcam →
facial morph + head + full-body skeletal animation, epic #869). This standalone
repo is for discoverability; the app fetches from the aggregate repo.