Tensorflow library isn't in the firmware after the procedures to make it

OpenMV Products
1

I made the framework as it is explained in this link:

https://wiki.st.com/stm32mcu/wiki/AI:How_to_add_AI_model_to_OpenMV_ecosystem

Now, I have a problem. The tensorflow library has not been added to the firmware, when I made it. Because when I run this code “tf_face_recognition.py”, I receive the following error on this line “net = tf.load(“trained.tflite”, load_to_fb=True)”.

OSError: This function is unavailable on your OpenMV Cam.

Then I updated the firmware to the release version and there wasn’t any error.

Can you tell me what should I do when I make the firmware to add tensorflow library to the firmware?

2

What target platform are you using? We don’t enable it in every MCU.

3

I’m using an OpenMV Cam H7 Plus. So, you know what the MCU is.
But if you need it, I can take a picture of it and send for you.

4

You can use TF and our examples or CUEBAI and their examples, you can’t use both at the same time.

5

Thank you, but it is too bad, because I created my network after learning it. Now, I want to load it with TF. But when CUBEAI can load it and give me the correct results, then it would be enough, but in other case, that would be very bad, as I thought, I can learn my network and use it on Openmv.
I added my network to firmware and received the output from CUBEAI, then there was an error in the output list from CUBEAI. That means, when I wanted to print this output list, Openmv restarted.
Currently, I created a smaller network and want to try it right now. Maybe, I can get the output from CUBEAI which has no errors in this output list.

Can you also tell me how to upload the firmware on the external flash? Because I want to use more functions and classes from Tensorflow, so I might need more space.
If you have these information on the website or anywhere else, please let me know about it. I searched in google, but cannot find.

6

Yeeeeeeeeees, when I reduced the size of the network, there was no error in the output list of CUBEAI, but now I must see that these values are correct or not. But at least, I know there are some bugs somewhere in the CUBEAI or other codes which I must find later.
But my professor want to buy a camera more than 2000 Euros which is special for neural network. Hopefully, there are not these bugs in that one.

After two week, now I know what the problem was.