Is there a way to convert a Roboflow API-trained model into a format compatible with the Sony IMX500 camera, which supports on-device AI?
What’s the best option for cellular data for the Pi? SIM card, USB dongle, portable router, etc.?
Hey @David293712,
I am unfortunately not aware of any methods to convert alternative object-recognition APIs onto the IMX500, and couldn’t find much info about it online. Perhaps someone else can offer some insight into this.
Regarding cellular data for the Pi, one of the most popular methods is to use a SIM card on a board such as the Waveshare 4G HAT. There are alternative options depending on the specific Pi you’re using.
Hope this helps!
Hey @David293712
Sony has a guide on how to do that, but it is quite involved, to say the least. You are also going to need some proper hardware or it may take a fair while. With an RTX 4070ti, it can take a few hours to convert a model like this.
Also, keep in mind that the AI camera is limited in the amount of memory it has (and hence the model size it can run). It’s hard to get a proper number but it’s around 8 Mb, so make sure your model is going to be smaller than that. I am unsure if I am remembering this correctly but I think that the conversion method does compress the model a tad!
