This repo contains an example inference flutter app that ports a CNN image classifier to Flutter on device.
For this purpose, we convert Googles mobilenet, as it is small with only ~3.5M parameters and runs with <200ms on inference on most modern phones and recognizes 1.000 classes.
To convert your model, or any HF model, to the onnx format, see the python-notebook model_to_onnx/convert_model.ipynb.
Run the flutter application with flutter run after installing all dependencies