Spleeter: [Discussion] How can I convert model to CoreML?

[kno3a87]

I wanna make CoreML.
I make .pb file from checkpoint, but I don't understand converting CoreML.
Please teach me🙇🏻‍♀️

[xk7155148]

I wanna make CoreML.
I make .pb file from checkpoint, but I don't understand converting CoreML.
Please teach me🙇🏻‍♀️
https://github.com/tf-coreml/tf-coreml
これを使うと、多分出来る

[kno3a87]

I saw the _Tensorflow conversion section in the Neural network guide_ and tried _TensorFlow 1 converter_ from this website, but I can't...

import tfcoreml as tf_converter
tf_converter.convert(tf_model_path = 'Protobufs/Binary_Protobuf.pb',
                     mlmodel_path = 'separate_model.mlmodel',
                     output_feature_names = ['softmax:0'],
                     input_name_shape_dict = {'conv_n_filters' : [16, 32, 64, 128, 256, 512]},
                     minimum_ios_deployment_target='12')    

I don't know what to write "output_feature_names", "input_name_shape_dict" and "minimum_ios_deployment_target", so I wrote somehow...
First of all, the way to make .pb file may be different...😭

[xk7155148]

I try to make coreml library under Windows.
Because I don't have a linux or OSX environment.😭
I can try to solve this problem, but it take some time.

[kno3a87]

Thank you for your doing!!
Please let me know if it takes time...!

[mmoussallam]

Hi there

It's not something that we have tried so far but you can have a look at the conversation in #155 to get inspiration.

Good luck

[kno3a87]

I saw it, but I cannot...

I saw the Tensorflow conversion section in the Neural network guide and tried TensorFlow 1 converter from this website, but I can't...

import tfcoreml as tf_converter
tf_converter.convert(tf_model_path = 'Protobufs/Binary_Protobuf.pb',
                    mlmodel_path = 'separate_model.mlmodel',
                    output_feature_names = ['softmax:0'],
                    input_name_shape_dict = {'conv_n_filters' : [16, 32, 64, 128, 256, 512]},
                    minimum_ios_deployment_target='12')    

I don't know what to write "output_feature_names", "input_name_shape_dict" and "minimum_ios_deployment_target", so I wrote somehow...
First of all, the way to make .pb file may be different...😭