How to convert model-large.tflite to model.tflite

[FireSon]

Hi I was trying out your solution with the image classification. And since I am new to the ML things I just followed all the steps in the juptyr notebook. After I finished all the steps it seemed that I got the same file as your model-large.ltlite and I was not able to get the model.lflite. Could you please help me understand how to create this file?
Probably there is some conversion taking place since the file is much smaller and also contains meta data. If I take the steps described to add meta data to the model my file stays about the same size and it does not seem to work quite as well on my android phone...
Cheers,
Jan

[jefure]

Hi, Jan
sorry for may late reply. I've never tested the model on android. So I'm not able to help you there. May there is a problem with the tensor flow version. I'm using the generated models to run with tensorflow light on a rasberry pi. Maybe the tutorial from google https://www.tensorflow.org/lite/examples/image_classification/overview can help you. I also used two different methods to generate the models. The first one is the method using the jupyter note book. This method creates a larger more performant model. You find the saved model in mounted output folder of the jupyter docker container. The second method uses the tensorflow model maker as described here https://www.tensorflow.org/lite/api_docs/python/tflite_model_maker. It uses the train.py script to create a simpler but smaller model. I hope this information helps
Kind regards
Jens

[FireSon]

Hi Jens,
No problem! In the end I solved it, although it was quite difficult since the versioning within python was quite sensitive. The model also works well on android. Do you still use your solution in real life? Does it work well?
Kind regards,
Jan