I'm currently working on an app using the ionic3 framework that recognizes hand-drawn characters. However, I am encountering difficulties with importing the model into my project. The model was initially imported from Keras and converted using tensorflowjs_converter. There are two methods through which I've tried to import the model into my ionic3 app:
- The
model.jsonand weight files (shards) are stored in the directory/assets/models. - The
model.jsonand weight files (shards) are hosted in Firebase storage.
When I launch the app in a browser using the first method, the model and weights load correctly, allowing me to make predictions successfully. However, when I deploy the app on my Android device using
ionic cordova run android --deviceBased on the provided shape, [3, 3, 32, 64], the tensor should have 18432 values but has 917To address this issue, I attempted hosting the files on Firebase storage. Despite fetching the model.json from storage, I encountered the same error as before, both in the browser and on the device.
My experience with storing shards and models locally led me to believe that the device does not recognize the shards in either method.
Furthermore, while using the Firebase storage approach on the device, attempting to fetch the model from the URL results in the error: Failed to fetch.
Below is the code snippet for retrieving the shards and model:
const modelURL: string = await this.db.getModel();
const shards: string[] = await this.db.getShards();
modelURL and shards contain download URLs from Firebase storage, with the model and shards located at the same level like so:
/* Firebase Storage hierarchy */
https://firebasestorage.googleapis.com/v0/b/project-foo.com/o/model%2Fmodel.json?alt=media&token=******
https://firebasestorage.googleapis.com/v0/b/project-foo.com/o/model%2Fgroup1-shard1of4?alt=media&token=******
https://firebasestorage.googleapis.com/v0/b/project-foo.com/o/model%2Fgroup1-shard2of4?alt=media&token=******
https://firebasestorage.googleapis.com/v0/b/project-foo.com/o/model%2Fgroup1-shard3of4?alt=media&token=******
https://firebasestorage.googleapis.com/v0/b/project-foo.com/o/model%2Fgroup1-shard4of4?alt=media&token=******
In order to fetch the model's download URL into tf.loadModel:
import * as tf from '@tensorflow/tfjs';
await tf.loadModel(modelURL).then(model => {
const output: any = model.predict(img);
});
Is there a way to pass the shards fetched from Firebase storage into tf.loadModel(), enabling me to retrieve all necessary data for making predictions in both the browser and on the device?
Thank you for your assistance.