Have you seen our fruit demo? If not, ​@David explains it amazingly here.

Basically, it is running 3 different yolov8 models on a single chip to process 4k camera input to segment fruit, while only using 640x640 models!

I love this demo and use it to evaluate any system that comes across my desk! So, obviously, when ​@Victor Labian posted the setup for the Orange Pi here, I just had to try it out! And it works perfectly straight after following the instructions!

I shaded the fruit to look different from the original to show the accuracy of the segmentation.

If you want to run this demo for yourself, you can! It’s available in the examples folder of your SDK, called fruit_demo.py. 

Let me know what you think!

Or, as ​@Spanner suggested

A Fruitful and A-peel-ing Orange Pi Experiment that Didn't Go Pear-Shaped

Here's what's current on the Axelera grapevine - a core functionality of the Orange Pi with Metis is detecting fruit in real time! Super smoothie AI pipe-lime with bananas and apples being picked from a live video feed. Turns out the Orange Pi has enough juice to easily run segmentation without going pear-shaped!