robertteleng/face-landmarks-coreml

FaceLandmarks · CoreML

Face landmark model converted to CoreML (.mlpackage) to run on the Apple Neural Engine, for use inside iOS apps.

Converted with robertteleng/coreml-forge:

uv run python scripts/export_face_landmarks.py

What this is, and what it is not

This is a format conversion, not a new model. The weights are MediaPipe's face landmark model, reached through MediaPipePyTorch. No retraining and no fine-tuning were done. Accuracy is whatever the original model has.

The value added here is the conversion itself: getting the graph to trace cleanly and produce an .mlpackage that the Neural Engine actually accepts.

Files

File	What it is
FaceLandmarks.mlpackage	CoreML package, ready to drop into an Xcode project

Limitations

• Not benchmarked on-device here: no latency or accuracy numbers are published with it.
• Inherits the biases and failure modes of the upstream MediaPipe model, including uneven performance across skin tones and under occlusion.
• Converted for a recent iOS deployment target; older targets may need re-exporting.

Licence and attribution

Apache-2.0, following the upstream MediaPipe model and MediaPipePyTorch. Credit for the model belongs to Google (MediaPipe) and to the MediaPipePyTorch authors; this repository only redistributes a converted format.