• Fix build of TensorFlow on raspberry pi. It seems like permissions issue re NSYNC? So should be easy to fix. [this is a real mess tbh]
• Write a guide for using pip to install pre-built TensorFlow binaries. Some of probs with this can be addressed by building protobuf on the pi. From what I recall it was NSYNC again for most of issues?
• Modify label_image.py to return variables we want and remove argparse etc. I.e. just write it again [completed]
• Code for detection devices needs to classify image using InceptionV3 graph first, then if we got >.70 elephant, go to use the graph from InceptionV3 retrained to classify specific class (e.g. lone, herd)
• Finish up with mounting in boxes [these waterproof boxes are proving difficult. I'm not very good at 3d printing
• Test as a wildlife detection camera trap. Upload everything we get on PIR==HIGH to twitter with labels and results. So should place it in woods []
• Test with horses [yes we can do this now, but we need waterproof boxes!]
• Test with cat-detector https://hackaday.io/project/20448-elephant-ai/log/69784-building-a-daytime-cat-detector-part-1 -- we are using this to test our supervised learning via twitter idea! [it's really bad for cats off-shelf w InceptionV3 so I am freezing this]
• Test with elephant images taken from Pi camera
• Go ahead with https://hackaday.io/project/20448-elephant-ai/log/68940-future-directions-and-improvements-to-elephantai so we need to a proof of concept with something local like birds in the garden for the supervised learning via twitter and virtual private machine. [underway]
• Try RF comms: give a detection device RF comms, and add a Pi with 2/3G + RF comm rx as base station. Try a repeater station too. So that would get RF comms from detection device and send it another few km to the base station! [ok, I am working on this per https://hackaday.io/project/20448-elephant-ai/log/70068-rf-network ]
• Make a Raspberry Pi HAT for the detection devices. So we don't want breadboards and jumper wires all over the place! So I'm going to try and make a HAT instead. This should be fun!

Well, I will update as I complete these!