Porting FastAI library for Android Apps

Hey guys!

I have gone through Part 1 of Jeremy’s Deep learning Course and have tried the lessons using FastAI library.

I was wondering if there was a way I could use the Image classifier which I have already trained and saved on the Jupyter notebook in the code of an Android App I am developing. Is there a way I can load this classifier in Android or port it in some way so that I can just use it directly in the code and predict rather than send API requests back and forth.

I have tried googling on this issue but I have found limited help. Any help or ideas on this matter would be really appreciated :slightly_smiling_face:

Thank you!

5 Likes

I have not done this myself but you can convert the FastAI model into Pytorch model and then convert the Pytorch model to a Tensorflow you can then use that model for android development, If you want to use Tensorflow lite you can easily convert Tensorflow model into a Tensorflow lite
Here are some that will help you achieve what you are looking for:
Fastai to Pytorch:
Try this first: Saving fastai model as a pytorch model
If the above one did not worked then see the link below:
https://mc.ai/convert-fast-ai-trained-image-classification-model-to-ios-app-via-onnx-and-apple-core-ml/
The link goes through how to convert a FastAI model into a CoreML Model to use it for iOS development. Just following the instruction of getting the Pytorch model

Then PyTorch to Tensorflow:

4 Likes