Am I the only one who picked a local model over a cloud API for my dad's fishing log app, and actually got it running?
Spent 3 weekends going back and forth between a hosted vision API and a tiny open model on a $180 mini PC, and picked the local one so his pond photos never leave the house. It labels fish in about 2 seconds now, so what would you give up to keep everything offline?
What made the 2 second labeling speed good enough to stop shopping around? With pond photos you're probably fighting glare and weird angles, so I'm wondering if the tiny model misses fish your dad knows are there. Did he have to change how he takes the pics to make it work?