Python bindings for TF & TF-Lite

[jose_ruiz]

Hello,
We have the need to use python 3 bindings and Keras (either Keras or tf.keras) as the front end for TensorFlow and TensorFlow Lite on i.MX8QuadXPlus. From the i.MX Linux Release Notes (L4.19.35) I can see that Hardware Accelerated eIQ Machine Learning is supported on this SoC, however, there are no further details on what features are supported for each framework (TF, TF-lite, OpenCV) and on what version. There is also no mention of the Python bindings for TF or TF-lite, are these available? On what version (2.7, <=3.6 or >=3.7)?

Is there up to date documentation regarding the supported features for each SoC?

Thank you
Jose Ruiz

joseogandoj‌

[vanessa_maegima]

Hi Jose,

In L4.19.35 TfLite supports running inference on GPU:

From Release Notes, 8QXP supports this NN acceleration:

Keras is currently not supported.

Python3 bindings for TfLite and ArmNN will be available in future releases.

Thanks,
Vanessa

[jose_ruiz]

vanessamaegima‌ Thank you very much for your detailed response. Is there an expected release date for python 3 bindings? Do you know what version of python 3 will be used (<=3.7)?

Thank you

JM

[vanessa_maegima · Accepted solution]

Next release will be available by the end of this month, March 31st.

Internal image at this point uses Python 3.7.