Android (Capacitor) app with trained Tensorflow model in assets folder

Capacitor
1

I’ve built an app that uses the following code to load a trained Tensorflow model (simple classifier NN built in Python).

import * as tf from '@tensorflow/tfjs';
....
const model = await tf.loadLayersModel('./assets/models/model.json');

This works on the browser but not when testing on Android. The app is written in Vanilla JS.

In the Android Studio logcat I see the following error:

E  Unable to open asset URL: http://localhost/assets/models/model.json

I also added these in the AndroidManifest.xml:

    <uses-permission android:name="android.permission.READ_EXTERNAL_STORAGE"/>
    <uses-permission android:name="android.permission.WRITE_EXTERNAL_STORAGE" />

So far I’ve looked into using the @capacitor/filesystem package but I’m not sure how to replicate the model-loading behaviour of loadLayersModel.

How can I replicate the working behaviour of the app running on the browser, in Android?
I’d eventually tackle IOS too but one thing at a time :sweat_smile:.

Thanks!