I wanted to get the Google AIY Vision Kit when it came out at the end of last year but they were very popular and I wasn't able to get one. I recently got a notice that they had stock of the new version of the kit at the local Target store and I made sure to buy one that day. Target has started to get into selling more Maker level electronics. They also had the AIY Voice Kits and I heard that they will be selling Raspberry Pis soon.
Link to Vision Kit at Target: https://www.target.com/p/-/A-53417081. A great deal for $89.99.
The Vision Kit is set up to allow users to experiment with AI applied to image processing. This is something that I'm extremely interested in. The image processing engine is a Movidius chip located on a Vision "bonnet" card that plugs onto the Raspberry Pi GPIO connector. The Movidius chip allows the Raspberry Pi to run TensorFlow and do the image processing locally rather than using cloud resources. The kit comes with pre-programmed examples to do smile detection, object recognition, and facial recognition. You can also compile new models for the Vision Kit, so you can train and retrain models with TensorFlow in the cloud, and then deploy the trained models onto the Vision Kit.
The original version of the Vision Kit just included the Vision Bonnet, the cardboard container and some auxiliary hardware. The new version is a complete kit including a Raspberry Pi Zero WH, a Raspberry Pi camera and a pre-programmed micro SD card. The only thing you need to add is the power supply and some way to interface to the Raspberry Pi (you can configure it via an Android app and run headless).
Vision bonnet with camera mounted on RPi0 WH.
It took me about 30 minutes to assemble the kit and try out the examples.
Link to Vision Kit instructions: https://aiyprojects.withgoogle.com/vision/#meet-your-kit
It all works as advertised. The housing leaves a bit to be desired, but I noticed that NewLifeTech designed a 3D printed case on Thingiverse: https://www.thingiverse.com/thing:2888996/#collections.
I'm going to try to print one later.
Looking forward to compiling my own models and taking it for a spin.