Using Edge Impulse for training models

Hi, I would like to know is it allowed to use Edge Impulse to train a model for OpenMV camera using my own data?
Thank you for your time.

can someone please answer my question so I know how to continue developing my robot?


sorry for not answering earlier. Your question is more complicated than it seems, since tools like Edge Impulse emerged pretty recently. We’ve had teams using similar software, I believe two years ago. It is an issue we are yet to discuss within a committee to determine how to regulate it, and if to forbid it. Therefore, take my answer with a grain of salt and I will try to provide definite verdict asap.

In my opinion, it should not be allowed. From my limited knowledge of this software, I have came to these revelations. Even though you are required to create your own dataset, you have no part in creation of NN architecture, tunning of hyperparameters, even augmentation of the dataset. All of these, and much more, are important skills to learn. And you miss out on them when using this kind of software.

I recommend using python libraries such as pytorch or tensorflow. There should be plenty of tutorials and examples available online, where most of the code could be reused, yet it gives you an insight of what is required for a successful CNN training. At the end, you are left with the same CNN model file, like the one from Edge Impulse ( I’ve seen that it offers pytorch as a output model), but has much more learning value.

Hope this helps for now, but I will get back to you as promised,

Matej Novosad,
2023 committee