Is there a quantized float16 version of the face_mesh_landmark.tflite model file?

[zhanxuejie]

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

None

OS Platform and Distribution

android11, linux

Python Version

3.9

MediaPipe Model Maker version

No response

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

facial keypoint recognition

Describe the actual behavior

The existing face_mesh_landmark.tflite is in float32 format.

Describe the expected behaviour

Expected face_cesh_1andmark.tflie to have float16 format and saved'model format

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

face_mesh_landmark.tflite

Other info / Complete Logs

No response

[zhanxuejie]

Hi, I would like to confirm if the face_landmark.tflite provided by mediapipe is float16 or float32. Could you provide me with a file resource in saved_madel format? I would like to use it for quantization and verify the effects of float32, float16, and int8 on my application. Thank you in advance!

[kalyan2789g]

Hi @zhanxuejie, We appreciate your interest in MediaPipe! While we previously offered a legacy system for Face Mesh landmarks, it has been superseded by our more advanced Face Landmark Detection solution.

To ensure the best performance and access to the latest features, we recommend using the newest version of MediaPipe. You can find it here.

If you encounter any difficulties after migrating to the latest version, requesting you to please let us know. We're happy to assist you further.

Thanks,
@kalyan2789g

[zhanxuejie]

Hi @zhanxuejie, We appreciate your interest in MediaPipe! While we previously offered a legacy system for Face Mesh landmarks, it has been superseded by our more advanced Face Landmark Detection solution.

To ensure the best performance and access to the latest features, we recommend using the newest version of MediaPipe. You can find it here.

If you encounter any difficulties after migrating to the latest version, requesting you to please let us know. We're happy to assist you further.

Thanks, @kalyan2789g

@kalyan2789g hi, thank you very much for your answer.

The latest solution I see is to package the model file and post-processing operations into a. task file, and I noticed that the latest model is float16. I hope to deploy my application on mobile devices, so I need a. tflite model file. How can I obtain the latest. tflite model?

[kuaashish]

Hi @zhanxuejie,

Certainly, you can extract the TFLite model file from the .task file of the new Face Landmarker by running the following command:

unzip face_landmarker.task

Additionally, we would appreciate your confirmation on the status of this matter. If it has been resolved, may we proceed with marking the issue as closed?

Thank you.

[github-actions]

This issue has been marked stale because it has no recent activity since 7 days. It will be closed if no further activity occurs. Thank you.