pycoral VS edge-tpu-tiny-yolo

I’m sorry to hear that you’re having trouble finding a suitable version of PyCoral for macOS Ventura. Unfortunately, I couldn’t find any information about PyCoral support for macOS Ventura either. However, one possible solution could be to build and install PyCoral from source on your Mac using Python 3.9.x or downgrade Python to 3.8.x and install PyCoral again. You can find the PyCoral source code on GitHub. Frigate NVR is designed around the expectation that a Coral TPU is used to achieve very low inference speeds. Offloading TensorFlow to the Google Coral is an order of magnitude faster and will reduce your CPU load dramatically.

Frigate NVR is an awesome open source project that not only provides recording capabilities, but also object detection with very good performance utilizing Coral AI's EdgeTPU. It also integrates well with Home Assistant so it opens up the door to do basically whatever you want.

I’ve never heard of Google Coral but it’s really neat. https://coral.ai

I guess because it'll be supported longer? I don't use photos or moments, so there were zero reasons to upgrade. I would lose out of the box ability to use my zwave/zigbee stick and coral.ai stick without workarounds. Plex migration to DSM7 seemed like a pain also (with more drawbacks), but I moved my Plex install to Docker to prevent any future breaking from DSM updates

I checked all the official distributors listed on https://coral.ai/products/accelerator/ and Mouser has ~1300 in stock (https://www.mouser.com/ProductDetail/Coral/G950-01456-01) for MSRP of $59.99. There are some on Amazon marked up to $85.

Google Coral is an offline/private image processor and is pretty amazing. You can read about it here: https://coral.ai/

Have you considered a Google Coral?

'pip3 install https://github.com/google-coral/pycoral/releases/download/release-frogfish/tflite_runtime-2.5.0-cp37-cp37m-linux_armv7l.whl'

The accelerator requires a tensorflow model. Here is someone's example to maybe get you going. Looks like some of the ops in the net aren't supported on the TPU, so you have to replace them with alternatives.