Hello, I wanted to know which library should be used to be able to use TensorFlowLite Micro with Nicla Vision. I have tried multiple libraries (The official one, ArduTFLite by Chirale, etc), however, most of them fail because of "Board not supported" error or "Cortex 7" related errors.
Is it not possible to use TensorFlowLite Micro for development with Nicla Vision on Arduino IDE, or am I way off track, and there are other methods?
Thanks in advance.
It probably is possible, but is TensorFlow Lite actually supported for Arduino?
It never attracted much interest, and seems to have died some years ago, but questions or complaints about it still crop up every few months on this forum.
Very interesting, do you mean for the Nicla Vision specifically or in general? There is the official TensorFlow Lite Micro library: GitHub - tensorflow/tflite-micro-arduino-examples
The issue is that this library, along with many others, don't really work for Nicla Vision. I was able to get it to work with ESP32-CAM with minimal effort (the ESP32 board package had tflite micro library in it), but for some reason, when it comes to Nicla Vision, nothing really works.
Can I not deploy models on Nicla Vision and run inference, or am I going with the wrong workflow?
Then you need to fix the library, which is certainly "possible".
But what is the point of running TensorFlow on an MCU?
I am trying to deploy a model on the Nicla Vision, but I can't run inference without TensorFlow.
As for fixing the library, is it really doable? I thought since libraries support CORTEX 4 only and Nicla Vision has CORTEX 7, it is a lost hope.
If this project seems too difficult for your current skill set and choice of hardware, then change the approach and/or the expectations.
I see, thank you.
One final Question, do you think it would be easier to do what I am trying to do on OpenMV instead, it seems to have a built-in TensorFlow light micro library
You haven't said what you want to do.
If the project has something to do with pattern recognition in images, then a modern laptop with a camera and Python would be a vastly better choice than any MCU.
Basically, it is gaze estimation using TinyML, so I must use Nicla Vision and deploy the model on it.
It seems that you have a great deal of study and coding work ahead of you.
Good luck with the project!
Thank you.