how to use .pkl file to make a tensorflow lite model?

[MONIKA0307]

I have trained Facenet model and got the .pkl file. Now, I want to convert it to a tensorflow lite model, which as per my research requires protobuf file. This protobuf is not changed after training, it is of same size as it was. So, it is not an indicative of the trained model. How can I do this now

[wangdan0527]

I have same issue here. @MONIKA0307 Did you find a solution for that?

[bmachin]

Have you guys tried the suggestions in this post: https://stackoverflow.com/questions/50632152/tensorflow-convert-pb-file-to-tflite-using-python

You should be able to use the resulting frozen graph protobuf file you get using: https://github.com/davidsandberg/facenet/blob/master/src/freeze_graph.py

I hope this helps.

[MONIKA0307]

Hi,

Sorry for the late reply. Can you ask specifically what you want to know.

Warm regards, Dr. Monika

On Fri, 13 Sep 2019 at 12:22 AM, wangdan0527 notifications@github.com wrote:

I have same issue here. @MONIKA0307 https://github.com/MONIKA0307 Did you find a solution for that?

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/davidsandberg/facenet/issues/1053?email_source=notifications&email_token=AJLRHRZ4L263PGDYSVP5MBDQJKFYLA5CNFSM4H7I653KYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOD6S4KVI#issuecomment-530957653, or mute the thread https://github.com/notifications/unsubscribe-auth/AJLRHR4BX5RBYVTGK5X4XI3QJKFYLANCNFSM4H7I653A .

[wangdan0527]

Hi @bmachin @MONIKA0307

I tried your solution. But the tflite output is completely different from protobuf model output for the same input. It's about 23 - 24 times multiplied.

After that I tried to convert frozon model to coreml directly. But the coreml is not working on iOS app, the result array is empty.

Any ideas?