Extracting Edge Impulse FOMO training script to local Computer

Help
#1

I’ ve been working on FOMO model for a few weeks. So, I tried a few times to run the FOMO training script on my local computer, all the attempts are failed because of the Edge Impulse s unique training libraries. I had tried to get those libraries but I couldnt. So, anyone have an idea for me?

#2

Hi @MrAccelerated,

Which script are you referring to? Are you trying to run the .eim model file in the Edge Impulse runner, or are you attempting to compile the C++ SDK locally?

Also, what error messages are you seeing? These will help us troubleshoot the issue.

#3

Hi @shawn_edgeimpulse,

I tried the code in FOMO training advanced mode (keras model). Here is the lines in the code:

Ekran Alıntısı

If it is impossible, is there any way to run this or equivalent code?
Or is there any equivalent code or file?

#4

Hi @MrAccelerated,

We’re aware of this limitation and at the moment you can’t run FOMO locally as the libraries are only available within the studio environment.
This will change in the near future, you can read more in this port from our CTO: Replacing iPython notebook support (in favor of something better)

Aurelien

1 Like
#5

Thanks @aurel ! Can’t you say any date for this change?

#6

Expected by end of September.

1 Like
#8

Hi @aurel
Since we are almost at the end of October, I wondered if local training with FOMO is now possible or not. If it is, where can I reach the local training script or other necessary docs to train a model?
Thanks in advance for your time!

#9

Hello @MrAccelerated,

We released end of September a feature called "Bring Your Own Model).
You can edit the existing blocks locally and test them too.

You can have a look at Bring your own model - Edge Impulse Documentation and GitHub - edgeimpulse/example-custom-ml-block-keras: Custom Keras ML block example for Edge Impulse

Best,

Louis

#10

How can I get the model structure of the fomo model

#11

Hi @w13www,

From the dashboard of a project that uses FOMO, you can download the .tflite model file. From there, upload that file to https://netron.app/ to visualize the structure.

I also go over the basic architecture in this video: tinyML Talks: Constrained Object Detection on Microcontrollers with FOMO - YouTube

#12

Hello @aurel and @louis, in the same question asking for MrAccelerated, local training with FOMO is now possible or not ?

#13

@aurel and @louis, i can run the model .lite out edgeimpulse ? Of example in raspberry pi4, just using a code ? And metrics abaout predict, with model h5 keras, tflite int8 ou float32 its possible run model.predict ?