Raspberry Pi have just released an AI camera module, in which AI image processing is offloaded to a chip integrated with the camera itself, rather than being performed on the Pi.
This sounds like it could be quite useful! This thread is to provide a home for any thoughts on how we could make use of it, and any modifications needed to accommodate it.
I had not seen that. The sensor size seems to be the same as the Pi HQ camera, which is a little large for the standard Openflexure optics without cropping. Only the Pi Camera 2 is supported by the Openflexure server v2, but the upcoming server v3 should allow other cameras, including this one.
Yes, and apart from the sensor size, it’s a little thicker than the non-AI camera module, so physical adjustments would need to be made to allow it to fit.
There’s a sensible article about it on TheRegister:
TheRegister: Raspberry Pi AI Camera takes inferencing load off the CPU
and documentation from Sony (who make it), here:
Sony: Develop with the Raspberry Pi AI Camera featuring IMX500
Apart from the ‘make it fit’ fundamentals, there’s then lots of scope* on the software side for programming it to do helpful things, that I haven’t even started to get my head around.
*(sorry)