I may have missed it in the documentation. Can I create a multi-tenant application with Edge Impulse? I will be using an Arduino Nano 33 BLE Sense board. Thanks.
Jan,
I want to use 2 sensors, each with their own NN connected together in a cascading architecture. If the first sensor’s inference comes back true, it fires the second sensor’s inference. If that comes back true, some action is taken. If either sensor’s inference comes back false, no action is taken. Does this help?
John
Ah, so yes, but it does require some engineering. If you look at the output of the Arduino or C++ library export the model is in the model-parameters
and tflite-model
folders, but these have fixed names / prefixes so if you add two models they are in conflict. You can get this to work by prefixing all declarations and variables in these with a prefix per model, but that does require some work unfortunately. We don’t have an easy way of supporting this out of the box at the moment
.
Hi @janjongboom,
I’ve been looking into this and I had a few questions:
- Are the only files that need to be changed in the
model-parameters
andtflite-model
folders? Or are there be files that have to be changed in theedge-impulse-sdk
folder? - Any declarations/variables/defines to look out for that might easily be forgotten while editing the names?
- Is it easier to combine two models that use the same sensor versus two models that use different sensors?
- Do you know of any projects that have succeeded running two models at the same time?
Thanks.
Hi @s-agar, in edge-impulse-sdk
you’ll need to be on the lookout for macros that start with EI_CLASSIFIER_
.
On 3) no, once this is resolved it shouldn’t matter which sensors you use.
On 4), not two fully separate models. Both anomaly detection + NN for example we see.
We have a backlog item to make this easier, but it’ll be a while until we have time to actually make this easier