Downloading individual FaceLandmarker models

[markpredigertrinamix]

Have I written custom code (as opposed to using a stock example script provided in MediaPipe)

None

OS Platform and Distribution

Embedded

MediaPipe Tasks SDK version

No response

Task name (e.g. Image classification, Gesture recognition etc.)

Face detection and landmarks

Programming Language and version (e.g. C++, Python, Java)

C++

Describe the actual behavior

We attempting to use Mediapipe face detection and landmarks on a custom embedded platform. We can run neural networks of usual formats like tensorflow(lite) or ONNX. However, the models available at https://developers.google.com/mediapipe/solutions/vision/face_landmarker are .task files, and we could't find any documentation of the file format, except the flatbuffer schema files.

Describe the expected behaviour

Where can we download the individual "FaceDetector" and "FaceMesh-V2" model files? Or can we extract them from the .task file? 

For the lecacy pipeline, model files are available at https://github.com/google/mediapipe/blob/master/docs/solutions/models.md.

Standalone code/steps you may have used to try to get what you need

-

Other info / Complete Logs

No response

[kuaashish]

@markpredigertrinamix,

Certainly, you have the capability to extract the FaceDetector TFLite model file and the FaceMesh-V2 model file from the facelandmarker.task task file by downloading through the guide here and executing the following command:

unzip face_landmarker.task

Additionally, we kindly request confirmation regarding the status of this matter from your perspective. If it has been successfully addressed, may we proceed to mark this issue as resolved and formally close it? Thank you

[markpredigertrinamix]

Thanks! I would have found it very beneficial it it were mentioned somethere in the docs that a .task file is just a zip file.

[google-ml-butler[bot]]

Are you satisfied with the resolution of your issue? Yes No