Hi, thank you for reading, I realize the issue might be due to an update to the YOLOv8 model or PyTorch onnx conversion output, see issue on ultralytics: https://github.com/ultralytics/ultralytics/issues/2317

I am more lean towards using PyTorch mobile or ONNX model at mobile device if no particular constraint on the need of using TF Lite framework as the conversion is a very painful experience

My example was run with:
-PyTorch 1.12.1
-onnx2tf 1.8.5
-ultralytic (from git as I need to disable mps in code, 8.0.54)