Convert to Core ML

[tucan9389]

[tucan9389]

https://apple.github.io/coremltools/

coremltools 2.0.0 does not support tensorflow 2.0 yet. 😟

coremltools has the following dependencies:

- numpy (1.12.1+)
- protobuf (3.1.0+)

In addition, it has the following soft dependencies that are only needed when you are converting models of these formats:

- Keras (1.2.2, 2.0.4+) with Tensorflow (1.0.x, 1.1.x)
- Xgboost (0.6+)
- scikit-learn (0.15+)
- libSVM

Alternatively, I'll port to pure Keras project, and then train and convert the model to Core ML model.

[tucan9389]

Related issue on Sep 2, 2018 (not solved): https://github.com/apple/coremltools/issues/231

[AndriiTsok]

Hi @tucan9389, have you tried tf2/tf.keras -> ONNX -> CoreML conversion?

[tucan9389]

@AndriiTsok No, did you try in that way? or others?

[tucan9389]

ValueError: Keras layer '<class 'tensorflow.python.keras._impl.keras.engine.topology.InputLayer'>' not supported.

The issue was solved replacing TF-2.0-alpha0 to TF-1.13.1.

[abrowne2]

Hi @tucan9389, did the conversion from tf2/tf.keras -> ONNX -> CoreML work when you replaced to TF-1.13.1?

EDIT: Found your 0.1 release containing release/cpm_model/model.* Are these the models from the PoseEstimation-CoreML repository, or this one?

[tucan9389]

@abrowne2 0.1 version is just cloned from edvardHua/PoseEstimationForMobile repostory. It's same with PoseEstimation-CoreML.

[tucan9389]

TODO

• tf.keras model(.h5) → frozen pb(.pb) → coreml model(.mlmodel)

Related issue

• https://github.com/tf-coreml/tf-coreml/issues/305

[tucan9389]

https://github.com/apple/coremltools/issues/446

[idchlife]

Are there any news on converting .pb to .mlmodel file?

[tucan9389]

Are there any news on converting .pb to .mlmodel file?

No in this repo, but the coremltools’s related issue was closed! I’ll try again.

[tucan9389]

TF2's saved_model is only supported on iOS 13 (minimum_ios_deployment_target=13) now.