Other hardware targets

[ebgoldstein]

• RPi Zero 2
• Google coral dev board mini + coral autofocus camera + some screen (or tiny OLED on Coral envt sensing board)
• nvidia jetson nano

Currently the model needs to output FP numbers in each of the outputs -- this means no TFLite micro targets. A work around is to make a model for % size classes: (i.e., % silt, sand (v. fine, fine, medium, coarse, v. coarse), v. fine gravel). this type of model could be done with a micropython TFL micro target, such as: openMV, portenta (black and white camera would be fine), nicla vision, Sony Spresense etc...