Pro tip: I switched from TensorFlow to PyTorch for my bird ID model, and my phone stopped crashing
My Pixel 6 kept crashing when I ran my bird call classifier on it because the TensorFlow Lite model was too heavy for the phone's RAM. I spent 3 weeks rebuilding everything in PyTorch Mobile, and now the model runs in 80ms without any crashes, plus I can actually train on my laptop instead of renting a cloud GPU. The new version missed only 4 out of 60 test calls at the park near my house, so I'm pretty happy with it. Has anyone else switched frameworks mid project like this, and did you end up regretting it later?